26:39Bryan Ng · Tutorial
How to Grow on YouTube So Fast It Feels Almost Unfair
A 26-minute live masterclass breaking down the three moves that turn a business owner's existing work into YouTube content that actually gets watched.
June 14thThe argument in one line.
A YouTube channel is seven separate systems, and AI only compounds results when each system is planned in granular detail before any of it is handed over to automation.
Read if. Skip if.
READ IF YOU ARE…
- You want to launch a faceless or AI-avatar channel and need the full sequence, not another isolated tip about hooks or thumbnails.
- You already run a channel and want to replace manual steps like idea research, script drafting and uploading with repeatable AI workflows.
- You use Claude Code and want to see someone plan a real multi-tool pipeline out loud before writing a single prompt.
- You need a thumbnail framework you can apply without a designer, including the five curiosity-gap categories and the A/B testing rules.
- You are deciding whether a channel aimed at a 55-plus, tier-one audience is worth building, and want the RPM math behind that choice.
SKIP IF…
- You want organic personal-brand growth on camera. Most of the tactics here assume a synthetic narrator and a volume posting cadence.
- You object to AI-generated channels on principle. The entire course is built around cloning a voice, a face and a format that already works elsewhere.
- You need vendor-neutral advice. The tool that carries most of the demos is one the instructor co-owns, and it is pitched repeatedly throughout.
- You are looking for verified earnings. The revenue figures come from third-party estimators and public product pages, not from disclosed statements.
The full version, fast.
The argument is that AI has collapsed the cost of making videos, so the remaining advantage sits in the systems around the video: who you choose to serve, how you hunt ideas, and how tightly you specify a pipeline before automating it. The build runs seven modules: pick an audience first (55-plus, tier-one countries, roughly an 8 dollar RPM) and let topics fall out of it; warm the channel and ship ten videos every two days to feed the algorithm data; run every video through a nine-stage board; pair a proven topic with a proven format borrowed from an adjacent niche; write scripts through a chain of deep research, verification, outline, ranked hook variations, repetition cuts and chunked humanizing rather than one prompt; engineer thumbnails around a stated curiosity gap and test them against your own trailing CTR; then hand rendering to a planned pipeline that costs roughly 20 to 30 dollars a video. The operating rule underneath all of it: spend 80 percent of the time planning and let execution run unattended.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:43
01 · Cold open and course map
States the seven-module promise, names the credential (50 million views, work for top creators), then applies a deliberate audience filter before starting.
01:43 – 07:41
02 · Step 1: audience before topic
A three-week-old AI-avatar channel with 148K subscribers is dissected as proof: AdSense estimated near 1,000 dollars a day, a 47 dollar ebook, and an 80-member community at 9 dollars a month. The lesson drawn is that the audience choice, not the topic, produced it.
07:41 – 12:49
03 · Three ways to find a niche
Mine Instagram and TikTok for avatar formats that have not reached long-form yet, port a working format to a different tier-one country, or attach a credible face to an existing faceless niche where trust matters.
12:49 – 17:05
04 · Wiring a YouTube data source into Claude
Walks through adding a custom connector in Claude settings so the model can query live channel and video data, then uses it to shortlist candidate niches and check whether they are currently monetizing.
17:05 – 20:30
05 · Step 2: trust score and posting cadence
Warm the account before the first upload, post a real test video, then commit to ten videos every two days. The stated reason is that YouTube needs a 20-day data window before it can place the channel.
20:30 – 29:54
06 · Branding package and five avatar rules
One prompt produces handles, description, banner prompts and a visual identity. Then the realism checklist: a real-looking background, human proportions on the rule of thirds, light in the eyes, imperfect skin texture, and no hands in frame.
29:54 – 42:43
07 · Step 3: the nine-stage board, built in Claude Code
Every video moves through idea, packaging, script, voiceover, visuals, editing, review, scheduled and published. The board is then built live in Claude Code, with most of the time spent forcing the model to ask questions until it is 95 percent certain.
42:43 – 48:41
08 · Step 4: three traps and the idea hierarchy
The traps are building the video before the packaging, spending five seconds on the idea and five hours on the edit, and making everything for existing subscribers. The correct order is idea, then title, then thumbnail, then video.
48:41 – 1:00:39
09 · Six places to hunt a proven format
Your own outliers first, then channels in your niche, adjacent niches with the same audience, larger niches you can borrow down from, unrelated niches where only the title ports, and whatever is top of mind this week.
1:00:39 – 1:06:38
10 · Mass-produce ideas, then filter hard
Generates dozens of data-backed ideas against live channel data, then ten title variations for the strongest one. The discipline is the funnel: a hundred ideas down to twenty, to five, to the one you actually make.
1:06:38 – 1:13:28
11 · Step 5: eight templates, angle and brand voice
Challenge, documentary, listicle, storytelling, educational, news, commentary and review, usually blended in a ratio rather than used pure. Topic, title, angle, template and brand voice all get defined before any writing starts.
1:13:28 – 1:28:50
12 · The seven-step script chain
Deep research through a research API, verify it, compile a report, outline with a reasoning-heavy model, write with a prose-heavy model, cut repetition, then humanize in 500-word chunks and run a final voice check. A client script in this chain took 30 minutes of compute.
1:28:50 – 1:42:30
13 · Step 6: the curiosity gap and five thumbnail categories
A thumbnail must stop the scroll and open a gap, and failing either loses the click. The five categories are moment, story, result, transformation and novelty, each worked through with real examples.
1:42:30 – 1:56:10
14 · Three C's, fifteen formats, ten principles
Content, composition and contrast as the design spine, then fifteen copyable layouts from title-headshot to motion-arrow, and ten principles covering relatability, the rule of three, congruence and visual hierarchy.
1:56:10 – 2:08:08
15 · Shooting expressions and A/B testing discipline
Eleven expressions shot on a phone at 4K 60, stills rather than video grabs, no coloured background light. Then the testing rules: compare against your own 24-hour, 7-day and 30-day CTR averages, test cold traffic, and change one variable at a time.
2:08:08 – 2:19:14
16 · Step 7: scoping the video pipeline
Plays a finished AI-narrator video, then maps the build backwards: script, voiceover, timestamped transcription, beat chunking, slide generation, screen recording, and rendering. Each stage gets a named tool and a documented fallback before any code is written.
2:19:14 – 2:33:38
17 · The autonomous loop and unknown unknowns
A single spoken brief sets up plan, execute, check, test and review, with a strong model planning and a cheap model executing, and a vision model watching the rendered output. The key instruction is asking the model to name the blockers it has not been told about.
2:33:38 – 2:38:01
18 · Recorded upload skill and channel intelligence
Demonstrates recording a screen workflow once and replaying it as a reusable upload skill, deliberately avoiding the capped and possibly penalised upload API. Then a competitor dashboard tracking 93 channels, and a retention audit against real AVD data.
2:38:01 – 2:39:43
19 · Close and calls to action
Recaps that subject expertise plus automation is the actual advantage, then points at the product link, the comment prompt and the next video.
Lines worth screenshotting.
- Pick the audience before the topic. A well-understood audience will consume almost any topic, but a good topic aimed at nobody in particular goes nowhere.
- Older, tier-one audiences are worth targeting because they watch to the end, tolerate simple editing, and carry an RPM near 8 dollars instead of 2 or 3.
- The best niche signal is a channel one or two months old already pulling large numbers, because it proves the niche is live right now rather than historically.
- Formats that work on Instagram and TikTok have not been ported to long-form YouTube yet, which makes short-form platforms a cheaper place to scout than YouTube itself.
- Changing the country is a repeatable idea engine. The same psychology that worked for a US audience usually works for the UK, Canada and Australia versions.
- Ship ten videos every two days rather than one perfect video, because YouTube needs enough data to work out who your channel is for before it can recommend it.
- The idea, the title and the thumbnail decide 80 to 90 percent of a video's outcome, so building the video first and packaging it last inverts the real hierarchy.
- Generate a hundred ideas, cut to twenty, cut to five, then commit to one. Prompting for the single winning idea skips the filtering that actually produces it.
- Eighty percent of content should target an audience you do not have yet. A channel that only serves existing subscribers stops growing at its current size.
- One prompt cannot write a competitive script. Deep research, verification, a compiled report, an outline, ranked hooks, repetition cuts and humanizing are seven distinct steps.
- Humanize a script in 500-word chunks, not all at once, because a long rewrite instruction gets dropped somewhere in the middle of a 4,000-word pass.
- Thumbnails have to clear two independent bars: stop the scroll visually, and open a curiosity gap. Clearing only one of them still loses the click.
- Thumbnail text should be additive rather than a restatement of the title, capped near five words, because repeating the title wastes the only space you have.
- The only meaningful CTR benchmark is your own channel's trailing 90-day average at the 24-hour, 7-day and 30-day marks. Industry averages tell you nothing.
- Test one variable at a time. Swapping the title and the thumbnail together makes the result uninterpretable and can kill a video that was already working.
- Ask the model to name its unknown unknowns before building anything you do not understand yourself, because you cannot prompt for a blocker you cannot see.
- Use the strongest model to plan and a cheaper model on low effort to execute, since planning quality determines the outcome and execution is mostly mechanical.
- A fully automated long-form video currently lands around 20 to 30 dollars in model and API cost, which is the real number to compare freelance editing against.
Seven systems, planned before anything gets automated.
THE FULL BUILD
The advantage is no longer in making videos faster, it is in choosing the audience, hunting the format, and specifying each pipeline stage in enough detail that automation has something real to execute.
02Step 1: audience before topic
- Choose the audience before the topic. A deeply understood audience will consume almost anything you make for them; a good topic aimed at nobody goes nowhere.
- Older tier-one viewers are worth targeting for concrete reasons: near 8 dollar RPM, higher completion rates, and tolerance for simpler editing.
- Treat a one-month-old channel already pulling large numbers as your proof the niche is live now, rather than studying a channel that succeeded two years ago.
03Three ways to find a niche
- The three reliable arbitrage moves are porting a short-form format to long-form, moving a working format to another tier-one country, and putting a face on a trust-dependent faceless niche.
04Wiring a YouTube data source into Claude
- Connecting a model to live platform data changes what it can do. Without it, you get plausible ideas; with it, you get ideas checked against what is currently working.
- Verify a niche by whether its channels are actually monetizing, not by whether the topic sounds interesting to you.
05Step 2: trust score and posting cadence
- Warm an account before its first upload and treat the first ten videos as a 20-day data-gathering run, not a launch you judge on video one.
- A consistent every-two-days cadence exists to give the algorithm enough signal to place your channel, which is a different goal from producing your best work.
06Branding package and five avatar rules
- Decide branding fast and refine later. Handles, descriptions and banners are a one-prompt problem that people routinely lose a week to.
- A generated presenter gives itself away in five specific places: the background, the proportions, the eyes, the skin texture, and the hands.
07Step 3: the nine-stage board, built in Claude Code
- Put every video on a board with one stage per column, because bottlenecks are invisible until work is forced to occupy exactly one state at a time.
- Spend most of your effort making the model ask questions until it is confident, then let execution run unattended. One-shot prompting is where these builds fail.
- Have the plan critiqued from several named expert perspectives before executing, because a first plan is a first draft regardless of who wrote it.
08Step 4: three traps and the idea hierarchy
- Build the idea, the title and the thumbnail before the video, since packaging decides most of the outcome and cannot be bolted on afterwards.
- Flip the time split. Five seconds on the idea and five hours on the edit is the most common and most expensive mistake in the hierarchy.
- Aim 80 percent of content at an audience you do not have yet, because serving only current subscribers holds a channel at its current size.
09Six places to hunt a proven format
- The reliable idea formula is a proven topic paired with a format proven somewhere else, since either one alone leaves you with no click or nothing to say.
- Adjacent niches with the same audience are the highest-yield hunting ground, because the underlying psychology transfers even when the subject does not.
- Borrowing a proven idea from a larger niche into a smaller one imports competitive pressure you did not have to survive.
10Mass-produce ideas, then filter hard
- Generate a hundred ideas, cut to twenty, cut to five, commit to one. The filtering is the work, not the generating.
- Ask for ten title variations on the idea you actually like, rather than accepting the phrasing that came out first.
11Step 5: eight templates, angle and brand voice
- Name your template blend and your angle before writing anything. Structure decisions made up front stop a draft from wandering.
- Brand voice is the limit on automation. A script that does not sound like you actively dilutes the channel, whatever it saved you.
12The seven-step script chain
- Replace one writing prompt with seven steps: research, verify, compile, outline, draft, cut repetition, humanize. Each step gets one job.
- Write and rank three to five hook variations instead of accepting the first one, and feed the model your best previous hooks as the reference set.
- Humanize in 500-word chunks, since a rewrite instruction reliably gets dropped somewhere in the middle of a long pass.
13Step 6: the curiosity gap and five thumbnail categories
- A thumbnail must clear two separate bars, stopping the scroll and opening a curiosity gap. Clearing only one of them still loses the click.
- The five gap types are moment, story, result, transformation and novelty, and naming which one you are using makes the design decision concrete.
- Optimise for the right viewer clicking, not the most viewers clicking, because attention that will never buy anything costs you retention.
14Three C's, fifteen formats, ten principles
- Design in order: content, then composition, then contrast. Deciding what is in the frame before where it sits prevents cluttered thumbnails.
- Cap thumbnail text near five words and make it additive to the title. Repeating the title wastes the only space you get.
- Run clarity, contrast and glance tests before upload, and if you cannot state the curiosity gap in one sentence, the thumbnail is not ready.
15Shooting expressions and A/B testing discipline
- One phone photo session at 4K 60 covering eleven expressions gives you a reusable asset library that beats pulling stills out of video forever.
- Benchmark CTR against your own trailing 90-day averages at 24 hours, 7 days and 30 days. There is no useful external baseline.
- Leave outperforming videos alone. The real risk in A/B testing is killing momentum on something that was already working.
- Test one variable at a time, thumbnail before title, on cold traffic rather than subscribers, or the result cannot be interpreted.
16Step 7: scoping the video pipeline
- Scope a pipeline stage by stage with a named tool and an explicit fallback for each one, before writing any prompt asking for the whole thing.
- Speaking a brief out loud captures reasoning that typing edits away, and the messy version gives the model more to work with than the tidy one.
17The autonomous loop and unknown unknowns
- Ask the model to enumerate its unknown unknowns before a build in a domain you do not understand, because you cannot specify around a blocker you cannot name.
- Route models by job: strongest on planning, cheapest on execution, and a vision model as the automated reviewer that closes the loop.
- Close the loop with plan, execute, check, test, review, and iterate until an automated grader passes it, so you are not the bottleneck on every attempt.
18Recorded upload skill and channel intelligence
- Recording a workflow once and replaying it as a skill can beat the official API when the API is rate-capped or suspected of penalising your reach.
- A competitor dashboard built from live data turns a vague sense of your niche into filterable evidence about what formats are currently working.
- Audit retention against real AVD numbers section by section, then fix the specific place where the script leaks, rather than rewriting everything.
Terms worth knowing.
- RPM
- Revenue per thousand views, after the platform takes its share. It varies by audience age, country and topic, which is why targeting a wealthier, older, English-speaking audience raises earnings on identical view counts.
- Tier-one country
- Shorthand for the US, Canada, UK and Australia. Advertisers bid more for these viewers, so videos watched there earn several times what the same video earns elsewhere.
- Outlier
- A video that substantially beat its own channel's average, usually expressed as a multiple such as one of ten. Outliers are treated as evidence of a repeatable format rather than luck.
- AVD
- Average view duration, the mean number of minutes a viewer actually watches. It is the retention number used to judge whether a script structure is holding attention.
- CTR
- Click-through rate, the share of people who click after seeing a thumbnail in their feed. It decays as a video ages, so it is only meaningful compared against the same channel's history.
- MCP
- Model Context Protocol, a standard way to connect an AI assistant to an outside data source or tool so it can query live data instead of relying on training memory.
- Curiosity gap
- The distance between what a viewer already knows and what they want to know. A thumbnail creates it by showing enough to raise a specific question and withholding the answer.
- Rule of three
- A composition guideline limiting a thumbnail to three main elements, so the image stays legible at mobile size and the core message survives a half-second glance.
- Faceless channel
- A channel published without a real on-camera host, typically using stock footage, generated imagery or a synthetic narrator, so production can scale without a person being filmed.
- Agentic workflow
- A chain of separate model calls where each step has one job and passes its output forward, instead of asking a single prompt to do research, structuring and writing all at once.
- Unknown unknowns
- Blockers you are neither aware of nor able to describe. Asking a model to enumerate them before a build surfaces failure modes you could not have prompted for.
- Deep research API
- A programmatic research endpoint that returns dozens of cited sources for a query, rather than the handful a chat interface surfaces, so a report can be compiled and then fact-checked.
- Test and compare
- YouTube Studio's native A/B feature, which rotates up to three thumbnails on the same video and reports which one earned more watch time.
Things they pointed at.
2:33:38toolClaude Co-Work record-a-skill
25:00toolChatGPT image generation, preferred over Nano Banana 2 for avatar faces
1:13:50toolPerplexity Deep Research API
1:14:00toolGoogle Deep Research API
1:16:00toolGPT 5.5 for outlining, Claude Sonnet 5 for drafting
2:14:06toolElevenLabs for voice cloning and voiceover
2:14:20toolOpenAI Whisper for word-level transcription
2:18:20toolFluid Voice, free open-source dictation used to brief Claude out loud
2:20:00toolGemini Flash used as an automated video reviewer
2:20:50toolKimi K3 and DeepSeek V4 as cheaper execution models
2:16:40linkawesomeskills.ai, source of the slide-designer and screen-recording skills
2:04:10toolYouTube Studio Test and Compare
02:34channelElias Yoder, the AI-avatar channel used as the opening case study
10:20channelOliver Pembroke, the UK pensions avatar channel used as the country-swap example
1:23:10channelFiction Profits Academy, client channel whose script is shown being written
Lines you could clip.
01:15
“If you can't sit and listen to this video for more than five minutes and you have the urge to go out there and watch Netflix instead, sorry to tell you this, but you're probably not gonna make it.”
04:30
“Ideally you want to start with the audience, not the topic. If you understand the audience on a very deep level, you can make any topic and they will consume it.”
29:55
“A calendar is just promise. A board is what keeps it.”
39:40
“You want to spend 80 percent of your time on the planning stage. Then on the execution stage, you can just let it run on its own.”
40:50
“There's this meme out there: hey, one shot this, Claude build me a 10K MRR business, don't make any mistakes. That is not what you want to do.”
43:40
“Creating your title and thumbnail all the way at the very end doesn't make any sense, especially since the title, thumbnail and video idea drive 80 to 90 percent of your video result.”
45:00
“Five seconds on the idea and five hours on the edit. You want to flip the ratio.”
56:40
“If I can take an idea that has performed very well in a bigger niche and apply it to a much smaller niche, it's like bringing a gun to a knife fight.”
1:02:40
“You want to generate 100 ideas, filter it down to the top 20, filter it down to the top 5, and then pick the one idea you have the most confidence in.”
1:19:20
“What you really want is to explain a point fully in the least number of words possible, and then cut off all of the repetition. That is what good writing looks like.”
1:21:00
“You want to humanize it in chunks. 500 words, 500 words, 500 words. This is the sauce.”
1:32:00
“There's only one winner, meaning that YouTube is a zero-sum game. When someone decides to watch one video, all the other videos lose out.”
1:47:30
“Creating thumbnails is not an art, it's engineering.”
2:03:30
“The only baseline for click-through rate is your channel's click-through rate.”
2:23:40
“Tell it to identify the unknown unknowns. When you're building something you don't already understand, you can't prompt for the blocker, because you don't know it exists.”
2:15:50
“You could go from idea to script to thumbnail to the full video being generated, and it will cost you about 20 to 30 dollars per video.”
2:38:30
“Once you combine subject matter expertise with knowing how to automate with AI, that is what really makes creating content effortless.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogystory
This is a full course on exactly how to start, scale, and monetize a YouTube channel using Claude. My name is Brian, and I've generated over 50 million views on YouTube and millions more dollars for the top creators on this platform. And now Claude has automated my entire YouTube creation process for my faceless channels, my personal brand, and my clients too.
So in this free three -hour course, I'm going to break down the entire process step -by -step, including how to use Claude to find the niche, set up your channel the right way, generate create viral video ideas, create an AI avatar clone of yourself, create high -click trade thumbnails in less than two minutes, edit full videos with just a few prompts, and then how to scale with analytics so that you can blow up a YouTube channel and monetize it quickly using Claude.
But wait, I have some bad news. If you can't sit and listen to this video for more than five minutes and you have the urge to go out there and watch Netflix instead, sorry to tell you this, but you're probably not gonna make it. And now that we have the real winners still here, let's get into it.
So step one is to find a viral niche. Finding a niche is the very foundation and it's the very first step. And if you completely...
screw up finding a niche, it's going to be way harder to grow your YouTube channel because you might not have the right audience and there might not be as much interest in your topic. So no matter how good of a video that you make, it's going to be virtually impossible for you to actually grow on YouTube. So that's why we have to start with this very foundation so that you can pick the right channel.
Throughout this video, the examples that I'm going to use is going to be pertaining to AI avatar YouTube channels. AI avatar YouTube channels is one of the biggest opportunities that I see out there right now to run. a clawed YouTube channel.
And I'm going to go into exactly why and how you can find the right audience. So for example, there is this YouTube channel, it's called Elias Yoder, and you can see it looks like a normal channel, but it has 148 ,000 subscribers, you know, 700 ,000 views on the very first video, 1 .3 million views on the second video. And you can see, look at the most popular videos, 1 .3 million views, 700K views, 500K views, 442K views.
And once again, this channel is only... three weeks old so it's less than a month old and it is 150 000 subscribers and if you were to look at analytics of this channel you can see that it's monetized first of all it has channel revenue that's not accurate data because you can see as 4 .7 million views assuming a three dollar rpm uh and if you look at the real time of the channel this is the number of views that they are currently getting it's 700 000 views every 48 hours so let's say it's 350 000 views every day that's a thousand dollars every single day just through AdSense, right?
And this is for one of their YouTube channels. This person I know runs multiple, multiple YouTube channels and it's doing about $1 ,000 a day, not including, if you look at their link description, you can see that there is a $47 ebook. People who watch their video, who also needs their problem solved inside of this niche, which is an Amish niche, absolutely printing like crazy because these are problem solving videos.
Whenever they sell an ebook, people who want their problem solved even more in depth will pay $47 for this ebook. And this is not the only way that they monetize. They monetize through AdSense, which is about $1 ,000 a day, according to NextLev.
And then they have an ebook over here. If you look at the post of their channel over here, you can see that they have a school community. And you see the school community has 80 people, $9 a month.
That's about $800 every single month, just in monthly recurring revenue. And this is something that is recurring, right? And you can see all of this is written of AI.
And I'm gonna show you exactly how to set this up for yourself, not just the YouTube channels, which is the very basis, right? You gotta get views in order to monetize these other ways. and then how to create digital products like these so that your YouTube channel can not only monetize through AdSense, but can actually monetize before the channel gets approved into the YouTube Partner Program.
And we're gonna go over all of that in today's video. So how do we find a niche? The first step is to pick the right audience first, and then the topic falls out of it.
And here's the whole system in one page. Ideally, you wanna start with the audience, not the topic. Because if you have the right audience and you understand the audience on a very, very deep...
level, you can actually make any topic and they will consume it. If you understand the audience, what are their specific pain points, right? Where are they from?
What is their age demographic? It's going to be a lot easier for you to make content surrounding them. Maybe one day their interest is like this Amish, another day they're interested in Mormons.
Their interest can change for a certain demographic or for a certain audience. But if you have a good grasp of who the audience is and you know exactly who you're targeting, kind of the topics will fall off. The topics are born out of their specific pain points or their specific interests.
As a general rule of thumb, we try to target an audience of 55 plus because these are older audience and older audiences not only watch videos for a longer amount of time, older people have learned in school and when they were growing up to finish one thing completely. They'll watch a video from start to finish. The average view duration on your videos will be substantially higher, right?
Assuming that your video is good. Obviously, if it's brain rot slop, no one's going to watch it in the very first place. If it's something that your audience wants to watch, ideally 55 plus, 65 plus audience, they are much more loyal to that one YouTube channel that they watch.
Also, as a rule of thumb, you want tier one audience, US, Canada, UK, Australia, which will give us $8 RPM, which is kind of the target here. Not like a $2, $3 RPM. If you target a niche, which has like older audience with like tier one audience, it's a lot higher probability that you have much higher RPM.
So you can see here, if you have channels that are with your audience graph skewed like this, this is very, very good because you can see 50 % of the audience. is roughly about 55 plus. And then there's very, very low audience in 18 to 24, 13 to 17.
In Canada or in the United States, all the demographics also have much more purchasing power because they have much more purchasing power. They are much more likely to purchase things from YouTube ads. Let's say you are YouTube as the platform.
How do you think about making money? YouTube basically makes money. Advertisers give money towards YouTube and then YouTube will take a percentage, 50%, and then give it to the creators.
If the viewers of these creators, buy products from watching these creators, YouTube is much more likely to say, Hey, these creators are doing a good job. Give them a thumbs up, actually push their content to more people, give them a higher RPM because advertisers keep coming back.
Advertisers will continue to increase their ad spend on that specific YouTube channel on that specific video, because you can see advertisers know that, Hey, we make more money when we run more ads. So we're going to run more ads. So they're going to give your specific video more RPM, which in turn means more money for.
your creators. You want to target a demographic with a high purchasing power. If you target kids content, if you target teenagers with this super duper high retention content, that's not going to end the best for you.
Audience is not going to be as loyal. The people who watch your content just won't quite be there. So once again, why does this audience win?
Because you have a higher RPM. There's more simpler editing because if you are older demographic, older audience, you don't really need all of these different flashy cuts. What you really need is a good hook and then the rest of the video to be simply cut, also have better retention because the viewer is much more likely to stick and stay.
Niches that are really good for these AI avatars, it will be like real estate or retirement and finance, maybe political news. Old people also watch a lot of news. Could be religious content or like doctors, right?
All of these are niches that pass tests because, you know, these are older audience that you're looking at. In terms of looking at like a real estate audience, right? Who are the people who are predominantly buying real estate?
It's not too many 20 to 30 year olds. 20 to 30 year olds are broke. So a lot of old...
demographic will watch real estate content. So let's talk about three ways to find a niche and then how we can automate it with AI. Number one is to mine IG and TikTok.
Instagram and TikTok is one of those very underrated platforms for finding these specific video ideas. And why do I believe that especially for AI avatars? Instagram and TikTok as platforms move a lot faster compared to YouTube because in a short form, you can produce these videos on an infinite cadence and therefore you see many more of these AI avatars.
pop up, probably 20, 30 times more AI avatars that are being created every single day on Instagram and TikTok compared to being on long form YouTube. And while the approach that most people will take is that they will go ahead and find these different AI avatars on YouTube and then try to replicate it, that is a very bad process, right?
Because it's already been done on YouTube. Why would you do it again? Whereas if you can find someone that is already on Instagram and TikTok and then port that over.
onto YouTube, you're going to find a lot more success. So if you can find AI avatars that are already crushing it on Instagram and TikTok, but not on YouTube, we can make the hypothesis a certain AI avatar has really worked on Instagram and TikTok. We can probably make sense that, oh, okay, it has a higher chance of it working on YouTube.
For example, there's this Instagram account, Mother Satori. You can see it has a million followers on this AI avatar. You can see all of this content is AI avatar, right?
This is an actual account. You guys can go ahead and search for it. They sell...
sour salt bitters here that's the product that they're selling from their instagram if you would have taken this avatar turn it into something similar that you make a youtube channel around it the chances of this succeeding will be a lot higher that's the very first method in order to find these channels and i'm going to talk a little bit more in depth into how we can do this with claude and number two it's to change the country what works in a tier one country will probably work in other countries we have seen this across so many different niches the car niche geography niche if there's a geography channel that's working well inside of the us there'll be one that works well inside of canada inside of the uk inside of australia right and you can take this exact same process for these ai avatar channels because once again when you change the country for a lot of these topics generally the psychology doesn't really change but the topics still apply and it's because it's a new market those topics tend to do very very well for example there's this channel over here oliver pembroke right hi i'm oliver pembroke a former chartered accountant specializing in pensions so you can see this channel
we will look at the most popular videos 10 days ago, 200 ,000 views. And once again, this is like an AI avatar that is talking UK retirement, right? All talking towards like UK seniors.
This is targeting older demographic audience doing very well because of this, this channel actually has very good potential for us to copy. It's because we know that this channel is working in the UK. We're going to want to see if that there are other channels that are also doing something similar inside of the US.
Maybe you can look at the Canada. Maybe you can look at Australia version, right? Those are all potential ideas that you can do.
right? And you can change the country of a channel that's working very, very well. Also changing the persona of who is working.
It's kind of like mixing both of these ideas, right? Mining IG and TikTok and then changing the country this way. And then number three would be to add a face.
It would be taking a faceless niche and putting a credible avatar over it. You can see this channel is actually fully AI. If you were to click into this, you wouldn't even tell that this channel is fully run by AI, but you know, it's posting once a day.
There's many, many videos that are breaking down channels like this, but you can see. this is like AI doctor. Okay.
So if you were to think where did the idea of AI doctor come from, there were a bunch of faceless channels talking about health and longevity. And if you can take all of those ideas from health and longevity, and then apply it into like an AI doctor, you attach a face to it on topics that require more trust. These AI doctors actually tend to perform pretty well.
And there was a whole wave of these AI doctors that performed very, very well. There's this guy, 122K subs. You can see this channel is monetized, you know, four months ago, 280.
27 ,000 views on his very first video. They also have a $27 ebook that they're selling, but you can see that you can add a face onto a faceless AI avatar channel, right? That will allow you to find a niche like that.
As a proof of concept, right? Look for proof from smaller recent channels, one or two months old and already pulling big numbers, right? 68k sub channel with a 1 .2 million view video will be a good example of a channel like that.
And that means the niche is live right now. And that way you can line up two options and send them to the team instead of guessing alone. So how can we do this process inside of Claude?
Let's find a niche that we can potentially do inside of Claude. Inside of Claude, there is one tool that I really like to use. It is actually called the subscriber MCP.
This is your AI system powered by content, right? So you can connect Claude, HBD, Cursor, or other AI tools into your subscriber account. And it will allow you to browse channels, research competitors, review scripts, and generate ideas without leaving your AI chat at all.
You can try a subscriber link in the description below, but this is like the one tool that I created. that makes Claude really, really OP because that way now Claude can actually talk to YouTube.
One of the big limitations of Claude in doing YouTube, if you don't have an MCP like this, has no ability to talk to YouTube. It isn't able to look at all of these different outliers. Whereas using an MCP like this, you can actually talk to Claude on a much more effective basis because you're able to browse YouTube channels and figure out what's working for a specific YouTube channel and then diagnosing it based on that.
I'm going to show you guys exactly how to set it up inside of your Claude account. You can just take this link over here, subscriber .ai. mcp subscriber copy it and then on the bottom left you can click on your name once you go into settings you can go into connectors and click customize and over here you'll be able to add your custom connectors if you click on the plus button over here and you add a custom connector you add a name this is subscriber mcp you paste the mcp server link within here and then you'll have your ooff client id which you'll be able to get inside of your subscriber account i already have it connected within this this is what it will immediately pull up you can see here it's a subscriber dot AI slash MCP.
And within this, you can see, you can write scripts, you can get your YouTube channel. You can get YouTube videos. You can also get channel ideas within this and get scripts, right?
You can add bookmarks. You can go ahead and generate ideas. All of it within this, we have this Intel feature and inside of this Intel feature, you can actually search YouTube channels and search YouTube videos and break them down that way.
And Intel is our database of 40 million plus YouTube videos that you can access through this specific MCP. And I'm going to show you exactly how to find a niche. with this.
So let's say, you know, the very first step is I'm trying to find a niche that kind of fits all of these different criterias that I just listed tier one audience. I'm 55 plus $8 RPM. I'm trying to think of, okay, like what niches are there?
So let me just ask for over 55 plus audience targeting tier one countries. You can see like life advice for older people, elderly wisdom guides, senior health tips, senior wisdom. This is all targeting older audience.
And it's actually able to find all of these Doug pest control, golden age of America. Okay. Timeless.
american history we could do like some kind of a historian right to tell all these different history nostalgia it's history over here 877k subs these are like historical kind of recounts you can see this is done of a human what if you can do some kind of like history kind of like documentary video that you can do with these AI avatars.
You can see these videos absolutely crush it. They're doing very, very well. What it recommends is senior finance and benefits, social security, SSA news.
This is the hottest one now. Retirement news hit 17K subs and only 83 videos. So it's actually able to pull in all of the different YouTube data and recommending ideas.
Untold stories of World War II. Untold stories of World War II averages 197K views per video. That's actually also not a bad idea.
Let's say I want to look into this. I want to dig into this niche. more find the top channels inside of this niche using subscriber for me so this actually uses subscriber tools in order to figure out okay like what could potentially be the best niches that actually match the audience for this a lot of clawed is like going back and forth once you have tools like this that we've pre -built it's inside the mcp once again you can click the link in the description to sign up and you know you guys can go ahead and test this out it's world war ii history enthusiasts and military you can see this channel is monetized 56k views you know kind of packaging is you know title thumbnail is not too too best here still getting good views so this is something potentially that i'll want to explore right you could do like ai avatar of person you know telling their personal stories on world war ii and then have images like that yeah like world war ii first person flavor is the
better entry right now. Nostalgia compilation is mature format. So let's do a World War II first person flavor.
Yeah, just like that, you know, going like back and forth, we literally went to like chat messages so that we could figure out a niche. Let's see, we're going to do a World War II first person AI avatar. Let's go on to the next step, which is the channel creation.
I'm going to talk a little bit about channel creation now that we have our niche, right? Because how you set up the first two weeks of your channels really decides whether YouTube will show your videos. And I'm going to give you the entire setup.
up within this. Step number one is to build a trust score. What you want to do before you upload a video is to actually warm up your channel.
So you don't want to dump content onto like a cold channel that you've never warmed up before. Before you actually go ahead and post a video, you want to make sure that you watch a couple of videos on that Gmail. Ideally, it's an older Gmail.
And that way, like YouTube knows that it's actually a real human and it's much more likely to actually recommend and push the video to other people. Secondly, you want to post a test video, one of like your real videos to see how that goes and then commit to a rollout. to the channel if the channel is actually healthy right when you're doing a test video on your very first video the metric that you want to see is you want to see that hey like impressions are actually going up in terms of the metrics that you're really really looking at when you are starting a new channel and you're doing your channel setup you want to ship 10 videos you know once every two days why do you want to do 10 videos ideally you do 10 videos you know once every two days that means it's 20 day testing period for a channel sometimes you never really know whether it's like your title thumbnail problem or your script problem and sometimes it just takes time for youtube to find your YouTube channel.
So just because you're not like the Elias Yoder guy and your channel didn't get a million views in the first five days, doesn't mean that you should give up. It actually means that you just need more time to give YouTube gathered data to train the algorithm and also improve your title thumbnails. The reason that you do it every two days is because if you do it with the system that I'm about to teach you, where Cloud really automates like 90 % of it, you can post consistently every two days, right?
You can go Monday, Wednesday, Friday, Saturday, and that way you can actually publish super duper consistently. This is just proof of all consistent. schedule and then once you've actually posted these 20 videos what you can actually do is to start reading your outliers once all of your 10 videos are out you'll be able to find the videos that broke out and work out exactly which one of them made them whim and actually double down on them and that's how fresh channel starts growing into a monetized one this is more about okay when you're setting up your channel you want to actually have a consistent posting cadence and with this consistent posting cadence you'll actually be able to determine you know what's working what's not working and then you're able to kill what's not working and then double down on what's working.
Doubling down is one of the most important YouTube concepts that I think you can appreciate once you've really grown a YouTube channel, because it's really, really hard to find viral ideas very, very consistently, right? Not every single YouTube channel, not even your top creators are as consistent at generating viral ideas.
There's a lot of creators that you have seen two or three years ago that had a big surgence and then fall down. You don't really know where they are now. You're not just trying to generate these viral videos.
They're trying to generate these viral videos on a very consistent basis. A lot of YouTube is a data game and it's looking exactly at what's working. So it's feeding YouTube enough data to decide exactly what YouTube should do with the videos.
And then YouTube deciding, okay, these videos are performing better. We kill the rest. And then we only double down on the ones that are working really well until it stops working.
And then once it stopped working, you go ahead and gather more data. You double down on exactly what's working that way. And that's exactly how you can.
and a fresh channel into a growing and monetized one. Now that we know how to warm up your channel and kind of the philosophy of growing your channel, I want to do and set up the minor channel details. And this is something that you shouldn't be so caught up in the details on.
You want to do it fast and you want to do it once and then you can refine it after. So that's why we're going to use Claw to do it in literally like less than five minutes. I'm going to put this prompt in here.
I want you to create a complete YouTube channel branding package for a new channel and you're going to give it your channel details. So you're going to give it the channel name, the niche, the target audience, what your main... topics are, what your tonal personality is, what your brand style is, and you're going to tell it to avoid the clutter.
So it's going to create your channel positioning for you. It's going to give you a channel name review. It's going to give you YouTube handle ideas.
Give me 10 handle options. It's going to give you your channel description. It's going to give you a banner text options.
It's going to give you a YouTube banner prompt, give you a profile picture, brand visual identity, thumbnail style guide, and then channel keywords, playlist ideas, your first 10 video ideas, and so on and so forth so that we can get started. Ask me questions so that we can kick. off.
this process right and your intern will get your internet to ask you questions so you can go ahead and train it to get the most outcome out of it what's the channel format it's going to be an ai avatar take the tone it's more of like documentary we can tell it it's going to be a world war ii narration storytelling so next and then what's the main channel goal adsense or reviews engine and then we're just going to let it process people really get paralyzed at the step what we want to do is we want to get through the step as fast as as possible, and that's why we're going to use Claude for this.
And really, while this is cooking, you want to think of Claude as your brainstorming buddy. Claude has limitations, right? And it's important to understand those limitations.
Some of those limitations include not having the ability to talk to YouTube, and that's why we use the subscriber MCP. Another limitation is that it doesn't always have all of the context that it needs in order to make a decision. We want to feed it as much context as possible.
One of those prompts that you can use is like, hey, ask me questions until you are 95 % certain about the output. and it'll keep asking you questions and then you can keep giving it input bit by bit. You can correct it and then it will actually go out there and do all of the work for you once it has sufficient context.
Bang, you can see over here, it creates a channel positioning, channel the last witness. That's pretty, pretty interesting. AI avatar narrator, documentary tone, AdSense views engine.
So it creates a channel positioning like this, channel name, last witness, letters from the front, a soldier's wall, wall diaries. I think all of these are like pretty good. So you can easily run if any of...
these description. We can just literally copy paste the description. And then it gives me these banner prompts.
So what I'm going to do is I'm just going to copy this to create a banner like this, just using Chachapiti image. You can use nano banana. You can use Google flow.
All of it works. So while that is loading, you can see it gives you a version B with no text. Profile picture.
You can use ChatGPT image. I'm just going to go ahead and copy this. Create this image.
I'm going to see if I like this logo that it comes up with. If not, I'm just going to take a screenshot of the avatar that I go ahead and create. And then it creates a brand visual identity, right?
It gives you the colors. It gives you a font style. This is literally something that would take hours and hours that you would previously create when you're working with a client.
All the parts with regards to the channel setup, we can just go ahead and do. right now. And before I go ahead and go into YouTube studio and actually create this, I'm going to go ahead and create the avatar first.
Let's talk about a couple avatar rules because we need the avatar in order to do the screenshot. And then we also need the avatar to go ahead and create the video. There are really five vectors that I see is like mistakes that people make whenever they're creating these avatars.
Number one, they don't have a realistic background. So if the background looks too AI, there's like too many AI books in the background. You can kind of see this background over here.
like super real number one it is real and then there's also like a dimension to it there's kind of like space between me and like the back and i can reach all the way back kind of like 3d and it's kind of popping out and you can also see real person whenever you're recording inside a professional studio what does it look like here i have my macbook here you'll notice that my eye is on the rule of three in terms of where it is on the camera right this is like the rule of thirds the person in the shot should lie within these lines and the eye line should be within this very third line there should be a realistic background you know kind of something with space here here you can also see that there's different variations in terms of the lighting uh and you want to replicate this within your avatar so having a realistic background not just some ai render having human believable proportions following the rule of three you can see kind of my face proportion this is like a human face proportion there's also life in the in their eyes right in a lot of these ai avatars kind of the main tell is inside of their eyes not just in the mouth movement but also in their eyes there must be some kind of a reflection in their eyes
when there's light that's being cast onto it. Having life in your eyes was typically pretty hard until recently. I'm gonna show you exactly how it can be done.
And then also number four, the skin texture being super hyper -realistic is one of the things that was pretty, pretty difficult. The problem with AI is that the skin texture is too perfect. If you were to look at my real skin here, as you can see that there's a lot of imperfections within this and you can see the light kind of bounces off a little bit differently on my face.
I'm just looking at myself over here. And then lastly, no hands showing. Typically where most AI breaks, eventually the technology will get.
there i don't want to zoom in too much or you'll not have the hand show maybe like your hands like that or maybe hands moving simply like that inside of the ai avatar this is example of a good ai avatar it does look like a real person you can see that there's life in their eyes you can see that there's actually some sort of reflection in their eyes he's kind of like tilted not fully sitting up straight you can see the shirt has like more creases so it's not like a super perfect shirt it does look like there's natural lighting you can see that there's lighting from these two sides and then there's also lightning coming in his face like this and also that side of the room is darker so the lighting on his face is actually right with this lamp it's kind of like zoomed out a little bit looks like he's having a professional camera these are like the main things that you'll want to take care of and i'm going to show you exactly how to use claw to create this so now i'm going to go back to this chat i'm going to paste in this prompt so this is a prompt that i typically use for creating these ai avatar images create a photo realistic ai avatar image for youtube talking head video format is this 1920 by 1080 identity lock use the same face character style expression
pose outfit background so that way i can go ahead and generate this and it's giving me this elderly veteran narrator it's not always going to be perfect in terms of very first generation so we're going to want to you know test it quite a bit i'm just going to copy this i'm not even going to look at the prompt and i'm going to paste it inside whatever image generator that we want you can see this logo isn't too good we can kind of see this banner last witness right every soldier had a story i think this is like super cool banner you can see i'm gonna go test these other prompts to see if there's anything that i really like that comes out of this it's very rarely that you're gonna get it out in one prompt right and once again you can use any image generator my preferred image generator here is gpt2 because i realized that the face looks a lot more hyper realistic and when it's a lot more hyper realistic it tends to look a lot more real compared to using something like nano banana 2 gpt is probably the best
in terms of the image model, kind of doing this Claude tutorial because I believe Claude is the best at automating a lot of these processes, especially ideas and script writing, which we will get to in a little bit. ChaiGBT is, I think, the most superior in terms of image generation right now, especially for AI avatars. I'll probably make an update video once that is no longer the case.
Man, you can see over here, this is the very first image that is generated. Actually, like this looks pretty good. Maybe like one of the things that I would notice, like his hair is a little bit too perfect.
maybe i'll go ahead and thin his hair and you know his suit does look professional there's some creases within this the lighting looks about right i don't really have notes on this specific avatar this one you know has more of the faults than the previous one you can kind of see that the skin is a little too perfect his hair looks a little bit too perfect his shirt is a little bit too perfect so this one's not as good and then also the background on this one does look a little bit more ai it's like okay who actually has a map and like a bunch of papers sitting around like that inside of their house in this very specific case okay maybe there's some like more papers here maybe i would remove this to the change of like some notebooks maybe the books look a little bit too ai so that's something that you can change but in general this is something that looks a lot better in terms of backgrounds like hey make the background more hyper realistic And bang, you can see over here, this looks way more realistic.
It's right. It's more pictures. It's him.
His hair looks a little bit more thinned out. So what I'm going to do is I'm going to copy this image. I'm going to paste it on this board over here.
Yeah. Now that we have this AI avatar, let me show you how I'm going to put together everything that I've talked about live, including the logo banner and the description here. So like over here, once again, we have a completely fresh tab in terms of the banner and I'm going to download this image.
literally just upload this as the banner the banner doesn't matter too much in terms of overall channel performance but it's something that's a good touch so that we can get it out of the way last witness and then in terms of the profile picture i'm just going to go on to this page i'm going to zoom into this and i'm going to screenshot the guy's face and then we can use the guy's face as the profile picture we just want something simple and done and then if i go back to my cloud i can see i'm going to check at the last witness can put this over here i'm gonna go to the channel description i'm just gonna take half of this bang You can add your email address within that.
You just click publish. And then that is done, which has allowed us to lay the foundation for the next couple steps. Now that we have gone ahead and generated the avatar, the very next step is actually putting together content systems before we go ahead and generate the video.
So now , right? A board is what keeps it, right? And this is what turns a channel into a content operation that actually ships on schedule.
You want to think about every video through these nine stages, okay? It starts with your idea. then your packaging, your script, voiceover, visuals, editing, review, scheduled, and then publish.
These are the exact nine steps that every single video goes through. And now before I go into the actual video creation, I want to talk about... all of the steps that goes into creating this system and managing it.
And now with Claude, you can very easily create all of these steps within that, you know, many, many videos inside of the pipeline. Let's say you run one channel, you have eight to 10 videos. It's good to be able to visualize where it is on each part of this stage.
And when you create a system like this, you get visibility as to where each video is. You can spot the bottlenecks when each column piles up. You can delegate it.
Every card is its owner and never question whose job it is. And then you can create a real pipeline. I'm going to talk about two ways that you can go ahead and build the board.
And I'm going to show you the board that I built first, which actually this second step, which is inside of the subscriber Kanban. And I'm going to show you exactly how you can build this for yourself based on your channel, because I know that, Hey, maybe your content system might be a little bit different from what I have, but this is what I personally use to run all of the different YouTube channels that I have.
I'm going to show you how I do it. And then I'm going to show you. exactly how you can do it and create one for yourself when you go into the subscriber dashboard you click on the channels let's say you click onto any channel within this you click on this very specific channel you can see that there are all of these different columns within this you have this new column you have this considering column you have this need script column script column thumbnail column video gen column video ready column and upload ready column you can create however many columns that you want but these are really the main columns that serve me right you can see that the video goes through eight different stages this is something that you know might team actually runs i don't even log in and look at this right now the new tab it can be for whatever use case that you want but inside of the new tab this is where i generate all of my new video ideas so let's say i click into this i have the description i have the angle for this that's all of it's created by ai and then i can also see the thumbnail inspiration for this video idea and then if i wanted i can click here and i can generate a bunch of different title options and once again this is all inside of my kanban board and i strongly recommend that you go ahead and actually build one of these for yourself
rather than going ahead and using something like Trello. Why build it yourself? Number one, it can actually be custom to your workflow and stuff like inside Trello and inside ClickUp, you're not able to write scripts or generate title variations inside of that.
You as the channel operator have a very good idea of all the different steps that it takes. to bring your video through. If you don't, you can copy exactly what I use.
This is what I use to grow all of my different faces channels and my personal brand channel. But you can, you know, have all of these integrated into your own AI workflows, into your own cloud so that you don't have to, you know, use all these external tools and pay for them. You can, you know, drag them through each stage.
It works exactly like a Trello board or a ClickUp board. You can, you know, write scripts within this. And so that way it's like an individual data center.
Previously, what was happening in this like pre -AI world is that I would have to pay for. click up, I would have to use Google Docs, all the SOPs would be everywhere. And now it's centralized inside one of my own Kanban boards because I can write scripts inside of this.
I can generate ideas inside of this. I can generate thumbnails inside of this. Really, it's one of the most effortless things.
I can just go ahead and manage everything within this. So let's say I have new video ideas that I generate goes into here, and I'm going to talk about how to generate new video ideas after. And then let's say three video ideas from this that I'm like, okay, I'm like kind of considering to actually do this, right?
And then I will have these. If any of these needs a script aka it needs to be published tomorrow or the day after i'm just going to drag it to the need script column so that i know that these need script and these are like the winning video ideas and this is where all of the all of the video ideas sit so i could have a hundred video ideas sitting within this and then considering i would have maybe two or three whenever i need more video ideas i'll drag more into the considering stage And then when I need something that needs a script, it will go ahead and create a script based on that.
And then now that I'm ready to actually go ahead and write the script, I'm going to drag it into the write a script tab. And let's say this script is ready and it's written. Now I need a thumbnail.
I'm going to drag it into the thumbnail tab. And then inside of this, I have my own thumbnail generator inside of this, which we can get into. a little bit later some juicy stuff within this and then once the thumbnail is ready i can just drag it into the video gen tab and then the video will be automatically generated which actually there's some cool technology around that i'll be able to show some of that to you and then once the video gen is ready you drag it to video ready and then the video is ready and you drag it to upload and then the video can finally be uploaded here and then there's interesting stuff like within this when it goes into upload tab you can click here and generate the description So it's even easy to upload.
You can see that I have built something like this so that it mirrors YouTube Studio. And because it mirrors YouTube Studio, the mistakes in terms of the descriptions being missing or timestamps that are missing is gonna be a lot less because I have it all inside one centralized database. I'm gonna show you how you can build this for yourself or you can just use subscribers tool.
And once again, you can click the link in the description to get this. And once again, the tags over here, you can click generate tags and it will generate all the tags relevant to your video because it has the script being generated inside. of this i can look at the script to see what tags that i want to generate and then i can go ahead and schedule this i can open youtube studio and do it within this i can also have a checklist that i go through all of these you can build custom for yourself so the very first thing that you need to do is to install the cloud desktop app so if you go to google download cloud desktop app or you can install this cloud desktop like really really easily and then you can click one of these downloads and then this will be inside of your system and it should pull up something exactly like the cloud browser tab but instead it looks something like this and now we in the Claude desktop app.
You can see over here on the left, loading the chats that we just had. But the main reason we want to have the Claude desktop app is so that you can use Claude code and Claude code is going to be one of those things that we're going to use to build these Kanban board and build these systems for ourselves. I'm going to go ahead and start a new session and then you can go ahead and create this Kanban board.
And really one of the best things about Claude is that really you can talk to it and it will be able to give you answers. So I'm going to let this work. I'm going to give it an initial prompt and I'm going to give you all of my different thought processes whenever I'm building a Kanban board like this.
I'm going to tell it, I want to create a completely new branch called Kanban demonstration. And I want to create a Kanban board for my video. pipeline for my ai avatar roughly i'm going to tell it like hey like these are like the different stages that i want so i want to have ideation stage script writing thumbnails video editing and upload ask me questions as to how this should work until you are 95 confident you know what i want let's refine the plan until you are 95 confident of the output So really one of the main things that I have learned from using Cloud Code for hours and hours and hours and really building these dashboards is that the most important thing is that you want to once again treat it as your intern, keep asking it questions and get it to ask you questions until you are very confident.
So let's see over here. I want this to be a standalone demo. How deeply should this integrate with the real app?
I just want to have a visual mock -up. Cards should be drag and drop. Should the board be scoped to one channel or span everything?
Let's say per channel. It's gonna keep asking me questions because you can see that the way that I've prompted it is that I'm telling, hey, ask. me questions until you are 95 % confident of what I want and then create a plan so that you are 95 % confident of the output.
The way that I think about these different planning stages, basically when you are creating something inside of Cloud Code or inside of Cloud, there is your planning stage and then there's your execution stage. And the reason the prompting stage doesn't really matter so much, you can actually get Cloud to ask you questions.
You basically want to spend more time on the planning stage. And then on the execution stage, you can just let it run on its own. Really, when you spend more time on the planning stage, right?
This is actually where 80 % of your work, right? Actually, you typing in the prompts and you giving it prompts should be right. It's, you know, creating this planning session, creating the architecture and an execution stage AI automates it.
You need to spend much more time on the planning so that it knows exactly what you want. And within this prompt, you can see that it's going to keep asking me questions. How should the demo page be reached and framed new routes?
What channel or topic should the sample cards be themed around? These should be themed around history channel. So it's going to keep asking me questions until it is like very certain of what I want.
And then it'll go ahead and execute that plan. What you're going to notice when I'm prompting this is that I'm going to tell it like, Hey, keep refining the plan. Hey, keep critiquing the plan.
And then this is where I will spend most of my time on. There are some cases that when I'm building something like this, I might spend an hour planning. This will allow it to actually execute this very specific plan.
Create all of these different agents that will go ahead and. execute this plan for me because it is very clear as to what I want in terms of this plan. There's this meme out there is, hey, like one shot this.
Hey, Claude built me a 10K MRR business. Don't make any mistakes. That is not what you want to do.
What you actually want to do is you want to give it all of this different context on exactly how you imagine it to work. Because let's say with this specific project, right, you are creating a content calendar. You have in mind like, oh, these are all of the different components that I want inside of my content calendar.
But you expect Claude to be able to read your mind. and be like, oh, obviously this is the way to do it. That is not necessarily the case, right?
Cloud has all these different ideas. It's trained on all these different projects. And there's so many different approaches that you can go ahead and create, for example, Kanban board with.
For you telling it like, hey, you copy how Trello does it. You're going to expect it to know exactly how that works. That's not really how it works.
That's not exactly how code works. Cloud is much easier to use. when you have that exact methodology of I'm going to spend 80 % of my time planning.
And then the rest of my time, you can see it's asking me so many questions. And then the rest of my time on execution, and I can go out and do other stuff while it's executing. Once again, right, this will be like more of a back and forth process.
Basically, I'm just waiting for it to gather all the context from me that it needs so that it can go ahead and do this next planning step. And then once I have the very first plan, I'm going to get it to, hey, refine and critique the plan. I'm going to get it to refine and critique the plan.
from a computer from like a senior devs perspectives and then it's also going to refine and critique this plan being a video expert or like a channel manager's perspective so that way you can give the ai all of these different roles and then it can go ahead and create all these agents so they can go ahead and refine and critique the plan And then really until you can like read the plan and be like 95 % certain that it knows exactly what it's talking about, then you can go ahead and execute it.
Only once it is so confident on crystal clear and asks you every single question on what it needs in order to execute it, then it goes onto the execution stage. It's come up with the plan. I'm not even going to read it.
I'm going to be like, hey, you are an expert channel manager. Critique this plan and tell me. all of the improvements you would make.
It's gonna go ahead, critique this on especially what is wrong with this, right? The very first plan that it comes up with, just like a human, when you create a first plan, let's say you are planning a YouTube video, it's not gonna be the perfect first plan, right? You're gonna do a second, third, fourth, fifth plan until you are very, very certain and you wanna think about talking to your cloud code whenever you are building out anything.
So we're gonna give this a little bit to go ahead and think, and then we're just gonna go back and forth onto this and until we can build something that actually. works bang and then you can see over here putting on my channel manager hat someone actually runs faces ai avatars through channels video editing is hiding four to five real distinct steps you're missing the real stages okay it recommends this and you can decide whether you actually want that my recommendation for the demo in this case i'm demoing it to you guys right research the 67 ai avatar yes yes rewrite this plan right so you can obviously give it feedback on like hey you don't like this plan or you like this plan it's going to go ahead and work on this plan individually right and it's going to give you output ask me questions until you are 95 certain of what i want right so i'm just going to queue this message so that it'll continue asking me questions until it's very clear like hey like this is exactly what i want from this and just like that once it's at 95 you can tell it go
and then it's gonna go ahead and build it so now once it goes let's go to the very next step which will be step four viral ideas and for this we're gonna do a little change of scene and Bang, we're back to my home. Let's continue with the rest of the video.
So you can see over here, this is like a simple Kanban board that was built, right? You can drag ideas from left to right. You go from ideas to the research or script, voice and avatar, editing and visuals with the packaging schedule published.
And this is literally all done with one prompt. You had just had to wait 10 to 15 minutes and you would get all of this. This is like an operational base, right?
You would have to do a little bit. This is very, very good. You click into this, you can see that you have upload, set title and description and screen cards.
pin common draft, schedule a slot, and you can see the title over here and the target publish. And you can have filters for all these different ideas that you have within this. It's pretty easy to put together.
And with just a little bit of refinement, you could really make this something like yours. I highly recommend that you build something like this if you don't already use subscriber. Yeah, let's get into the next part.
We will be talking a little bit about how to come up with viral video ideas because the idea is now the single biggest lever that you have to actually ripple your views with half of the work, especially with faceless channels. A lot of this stuff is going to be automated with AI right now.
Even the editing, the editing style can no longer be the mode. It is actually how good you are at coming up with video ideas, at understanding YouTube and understanding the data that you have at hand, and then deciding what to do with it. So these are the three traps that keep small channels small.
What a lot of people will do, especially if you're a beginner, is that you will think backwards. You will start with the video first and then go on. the title.
So most people will build the video and then they will build on their packaging. So there's a lot of people start off thinking that, Hey, I'm going to spend all of my time creating the video, making sure that the video is the best video possible. And then they go ahead, create their packaging.
This is a very backwards mistake because creating your title and thumbnail all the way at the very end doesn't make any sense, especially since the title thumbnail and video idea. drives 80 to 90 % of your video result. When you think of YouTube, YouTube is a click -first platform, meaning that when you go onto the YouTube platform, the very first thing is that people don't actually watch your video.
The very first thing that people see is your title and your thumbnail. Then they decide whether they want to click on it afterwards. When you think about the viewer's journey, you'll see when their eyes land on the YouTube homepage, the first thing that they see is your competitors.
So your video idea has to stand out from everyone else's. You wanna make sure. that your thumbnail and your title stands out from everyone else.
You can see when I'm looking at the YouTube homepage, very first thing that I will see the thumbnail. If the thumbnail is interesting enough, it will bring my eyes towards the title. And if the title and the thumbnail works together interestingly, and the video idea itself is very, very interesting, I will go ahead and click on the video and then watch the video.
The video idea and the way that you package it. is by far the most important thing that actually drives youtube results from a hierarchy in terms of how you go about putting your video together it always starts with the idea followed by the title and the thumbnail and then you go and create the video you want to have your idea your title and your thumbnail figure it out first before you go ahead and do anything else then the very next thing having a wrong time split it's spending five seconds on idea and then five hours on edit you want to make sure that you flip the ratio and then lastly is to only make your ideas for fans making your content for your current subscribers of course it is important to appease your video idea but you want to think about how you can not only appease your current audience but to also expand your audience your existing audience will be appealed by your more recent videos but you want to follow the 80 20 rule 80 rule is that you want to make sure 80 % of your content appeals to a new audience.
20 % of your content appeals to your current audience. This is the frame that you want to think about your video ideas, not just appealing to your current audience, because if all you do is appeal to your current audience, your audience won't actually grow. Whereas you want to find the right balance between appealing your current audience and growing a new audience.
In terms of the recipe for a video that actually grows your audience, This is by far the simplest way to figure out what video ideas to do. It is finding a topic that is viral, that is not only viral, but there is also cohesive with your brand, something that's on topic and pairing that with a viral format.
And that will give you a viral video. So the topic is what you talk about. Let's say it's an animal's channel.
It's life about an animal. Let's say the topic is a honey badger. And then the format is exactly how you present it to recipes of a viral idea.
In this case, the format is why it sucks. be born as if you don't have a good topic and you have a great format that doesn't matter right that won't actually give you better results if the topic is bad people don't want to even hear about it in the very first place but if you have a very great topic but your format is very bad that means you present it in a very poor way then no one is going to click on the video in the first place so that's why you need both a viral topic and a viral format so you want to steal a proven format from another niche and bolt it onto your topic and we're going to talk about how you can use claw to do that in a little bit So you can see over here, the topic is honey badgers, right?
You can see honey badgers don't care. There are so many videos on honey badgers that you wouldn't actually think gets millions and millions of views every single year. And it's something that's super consistent.
People, I guess, are just curious about what life is like as a honey badger. Format that we took, why it sucks to be born as a medieval executioner, right? Why it sucks to be born as is the format.
So you can go ahead and switch out that subject many, many times. And there's whole channels that are being built on this. So when you pair the topic in the format, is it the honey badgers is the topic you pair it with a format white sucks to be born as and then you get a viral video idea out of that the result in video 1 .5 million views this was over a year ago you can see the thumbnail is taken from the format too so you can see it's a honey badger and then it's arrow terrible life so how to use claude in order to do this the very first place that you want to hunt for proven formats is on your own channel so it's to read your outliers and find the videos that beat your average and run them back because you can see that these videos are already working on your channel so why not just double down on it this video on the full youtube script writing course my very first viral viral video you can see is 239k views over the course of two years so it's by far my best performing video since this did really really well on my channel why not i just redo the video literally a year later and you can see this is a four hour long video a beginner's guide to youtube script writing and this video is also one of my best performers 58 000 views right in this tiny tiny niche youtube script writing no one else would have made this
full YouTube script writing course video, even less would people have made it for full hours in length. And you can see, because I doubled down on a video idea that has already worked on my channel, this video was undoubtedly a one of 10 and it's still getting views to this day. It's my second most popular video.
You can see over here at 58K views. Once again, this is a tiny, tiny niche. You can see the views aren't super duper high because the interest in YouTube script writing isn't going to be as high as the interest within the business niche.
But inside of your channel, There are hidden gems already, the ideas that you are sleeping on that you can just immediately go ahead and... create.
The very first place that I will look at is within your channel, the ideas within your channel that you can already do that will perform very, very well. Number two is channels within your niche. It's looking at the top three channels, sorting by popular and then studying the top 10 of notepad and adapting with a twist.
I'm not saying to go ahead and copy someone else who has made a specific video. You want to put your own twist on it so that it is a little bit different on a idea level, on a packaging level. For example, on my channel, I posted this video literally a day ago, as of recording of this video, Cloud Code plus YouTube plus AI Avatars equals $42 ,349 every single month.
You can see this video is performing very well. It is a one of 10, right? And this is not an original idea.
I didn't come up with this idea. You know, this guy did this video over here, and this is also a Danny Y video. You can see this video has a million views in two weeks.
I was a little late to this specific trend. I put my own twist on it being Cloud Code plus YouTube plus AI Avatars equals, you know, $70 every single month. Thumbnail does also look a little bit different.
Even I hadn't posted a long time on this channel. It still performed very, very well. It's just putting my own twist on it because I wanted to talk about AI avatars.
It's something that I'm very, very passionate about. It's something that I work on every single day. And then now number three, it's looking at adjacent niches.
It's looking at similar, but not identical, abandoned places, urbex, forbidden places, and then the port winners over. How do I define adjacent niches? So adjacent niches, niches with the same audience, but slightly different interests.
This is. really really great way in order to find video ideas the idea hasn't already been done inside of your niche and if you can go hunt a niche with the same audience but with slightly different interests the psychology of the title and thumbnail and the idea are going to be very much identical and if the psychology is going to be very much identical then there's a high percentage chance that the specific video idea actually works let's say you're interested in facebook ads as your channel it's about facebook ads you can go ahead and look at Google ads niche or Pinterest ads.
If there are ideas that have worked inside of the Google ads niche or inside of the Pinterest ad niche, and you would have poured it over onto Facebook ads, it's actually a very high probability that specific video will do very, very well because you have found it in adjacent niches. Another example would be Canadian geography.
If you were to go ahead and look for Canadian geography channels, you can go ahead and find what are certain ideas that would work well in Canadian geography. You will look at US geography, Australian, geography, pay geography, you would be able to look at all these other niches.
I was like, hey, what are the topics within US geography, Australian geography, UK geography that could work well inside Canadian geography, roughly the same audience, right with the exact same psychographics. If an idea worked in US market, let's say you made a video idea on Texas, what it's like living in Texas, you could do similar video to like what it's like living in Canada, it could be a faceless channel, it could be a phase on channel, whatever channel you're running this principle.
applies the exact same if you are making a video about japan worked well to an audience who are interested in japanese culture maybe even make a video on korea exact same psychology works people who are interested in japan high likelihood they are interested in thailand interested in korean culture maybe interested in chinese culture and exactly how that economy works and that will allow you to create video ideas like that in terms of an example of looking at adjacent niches how to grow on youtube so fast it feels almost un thumbnail looks a little bit odd right there's like a guy with a camera and then there's youtube studio dashboard there's me on stage and then there's a bunch of people who are like looking at me uh one of the talks that i had at one of the events that we held over here in medellin colombia this was the first youtube event this was recorded in may and this is just me going over my talk and exactly the process that i had is with coming up with video ideas and growing on youtube i'm like hey how can i package this specific idea i looked at other people who did talks like this and actually repurposed their talk
Nick sitting, he made this specific video two weeks ago. It's him talking. It's him looking at a crowd.
And you can see over here, this is obviously not inside of the actual presentation that he's done, but he has Photoshopped over here two hours and he's kept everything else identical. He's not changed a single thing. And you can see this video did very, very well.
It was a one of eight for him. Clearly you can see it's the exact same format. Obviously it's with a slightly different audience, right?
This guy doesn't make videos about how to grow on YouTube, but you can see over here, it actually performed very, very well inside of his niche. It's exact identical format where it's him presenting in front of a stage. How can I take this thumbnail and replicate it for my specific channel and that specific event?
We had a photographer, we had a videographer. I was able to find this shot of me, you know, talking in front of the screen with like the camera on the side. So I took that specific shot, but I didn't want to show the exact same dashboard.
I wanted to show like, Hey, how to grow on YouTube. So it looked like I presented this slide inside of my presentation when in reality that's something that i just decided to photoshop out this was a very very easy way in order to come up with the specific thumbnail i stole the bracket full masterclass from this and i changed the title a little bit but this is a really great way in order to find inspiration from adjacent niches once again this topic it's not like the biggest biggest you know total addressable market topic so it's not going to go mega mega viral but you can see over here this is a one of 10 this performed exceedingly well for my specific youtube channel and my specific use case number four will be to look at bigger niches in my specific case youtube growth it's helping people implement ai implement best practices with youtube onto their youtube channel so that they can grow bigger niches is thinking about hey what's bigger than that let's say there's youtube script writing which is where my channel started out this is the box that i'm in if you were to really expand the box that you're in talking about youtube growth and then within youtube growth there is an even bigger bucket within this which is how to grow on social media and then within how to grow on social media you can see here that there will be different pockets
IG, TikTok. Okay, why do people want to grow on social media? Okay, people want to grow on social media because they want to learn how to grow a business.
Look at this YouTube growth niche, right? Which is, you know, kind of the videos that I make. And I could go and look at bigger niches for myself, like how to grow on social media.
You can see that that niche, how to grow on social media, is actually a bigger niche. There is a lot more competition and there is also a lot more interest. But if I can find a video idea on how to grow on social media that people really, really like, I can immediately...
take that idea of how to grow on social media and then pluck it into youtube growth that's actually a really really good way in order to find great ideas for youtube growth because the competition on how to grow social media is going to be so much higher there's going to be way more people making videos on that but if i can take that expertise an idea that has performed very well in a bigger niche applied to a much smaller niche it's like bringing a gun to a knife fight the idea has done really well in social media so we can make the hypothesis when i take that idea and i plug it onto youtube growth that will actually also do very very well that's also another method of finding video ideas for example when i am tracking my competitors i'm not only going to be looking at youtube growth and youtube script writing i'm also going to be looking at what are the ideas that are working on how to grow on social media what are the ideas that are working on how to grow a business and then i can plan my specific video ideas based on that that's also another really really great way to come up with great video ideas that way and once again i'm explaining the methodology so that we can go into claude and come up with video ideas a little bit after number five looking at a different niche this is a little bit unorthodox but i wanted to add all of the ideas here so that you can kind of have
a good overview of what exactly works. So looking at a different niche is also a really great way looking at entertainment, looking at gaming, but you won't necessarily want to copy the title. This is a little bit more rare in terms of porting it to your niche because sometimes audience within gaming, within entertainment will not have the same psychographics and will not think the exact same way as someone who's interested in business.
Same title thumbnail may not port over unless there is a good reason that you might think it actually has a good chance of porting over. One good way of porting it over. to another niche would be this specific video idea you can see this video idea was done two weeks ago 1 .2 million views once you see this you'll see competitive games this is very masterfully packaged you can see that hey actually there is potential once you see this you'll see youtube growth differently this is a really classic example of okay you just swap competitive games out for youtube growth and this is an idea that might work i don't think the thumbnail will specifically work for this so i would have to come up with another thumbnail idea because there's not necessarily something where you have intersection it's really important to take into consideration, does this specific thumbnail format really work?
Here, you can't really steal this specific thumbnail format, but you can probably steal this title and it would perform very, very well. And then lastly, it's presence of mind. It's always hunting what people are thinking about.
This latest video that I did, right, it's not only that it was a winning format within my niche, it also combined kind of like two techniques where there was this AI avatar, this guy who was performing very, very well, you know, zero to 200K subs in less than a month. there were a lot of people who were talking about this because there were a lot of people talking about this Elias Yoder character.
It is important to kind of capture the attention of people when it is immediately top of mind. People are like wondering, hey, how can I build a channel like this for myself? And that's why I have him included inside of the thumbnail.
This topic of this AI avatar was really, really trending and everyone was wondering how to make videos like this. If I were to go ahead and make a video on him and exactly how he has done it, that video would end up performing very, very good. It's just understanding what are the things that are top of mind.
for people and then you can make a video on that. It's not necessarily, okay, a lot of people are making videos on this, so I'm gonna make a video on this topic as well. It is understanding, hey, what is top of mind for people?
For example, right now, World Cup is happening, so a bunch of the World Cup videos would perform very, very well. Pretty much if you post anything related to the World Cup that's slightly different from the competition, your video would perform very, very well. Before you commit to any video idea, I always like to filter it through this.
You want to score your specific video idea out of 10 for each of these. my ideas through each of these like hey is there such demand like are people searching for this are people viewing content like this is there actual proven performance because if people don't actually want to view the topic even if there's proven performance there's no point of making the video 80 of your ideas should be appealing to a new audience and then 20 should be appealing to the existing audience so you want to make sure that your ideas are always expanding your audience rather than only feeding your existing audience let me show you how i can do everything live right now inside of cloud we're still gonna use the subscriber MCP.
It will be really efficient. Use it to help us hunt for adjacent niches, videos that are working inside of the niche in order to come up with video ideas. I'm going to go ahead, generate video ideas for my personal channel.
I'm just going to talk to it like a normal person. I want to come up with viral video ideas for my channel. I want to use the subscriber MCP.
Find me adjacent niches channels and what videos are performing well. of my niche bang you can see over here that it goes ahead and finds all these other channels has a couple like other script writing channels also finds all these other channels that are like ai you know youtube channel growth you can go ahead and open some of these channels to go see what they're about it told me this you have to read on your niche adjacent channels worth modeling and the video ideas that for all of it says okay it finds within my channel that okay these are the mega outliers 259k views 56k views and your strongest repeatable format was a creator economic study found all of these different channels for me, right?
And it gave me all these specific video ideas. So let's go through them one by one. If I started a YouTube channel in 2026, here's exactly what I would do.
Not a terrible idea. That's something that could be done. I scripted a thousand videos.
Here's the hook formula that never fails. This one's also pretty good because we can see that there's a bunch of the hook videos that have done very, very well inside of, you know, Callaway's channel hook is a topic is very, very trending. And then we can also add authority with that.
We can see on my channel that over here, some of these videos, I scripted a thousand YouTube videos. this is one of my top performers how pro youtubers use ai to write scripts and a new flagship course write a viral youtube script in 2026 when i already done it this one uh was a one of 10 didn't do as good as the others but it still did not too shabby how this creator made x dollars from one video and new case studies your iman video already validated this in your channel i think this could also you know potentially be interesting i don't know if there's anyone else that i'd already done this video for but you can see over here it is able to understand your youtube channel or it's able to understand you're able to point it towards it's a similar channel that already has videos or may not have that many videos and kind of have an idea of the audience.
And you're able to come up with video ideas like this. And what kind of was the conventional approach to YouTube? You would have to write down and you'll generate all these ideas slowly, slowly, and look at all these outliers, what's working in your niche, and then come up with ideas based on that.
That has become a lot simpler right now. You can generate ideas in volume. Based on this, I can be like, okay, it generated me six ideas.
I can go ahead and tell it like, Hey, I want. to come up with 20 more ideas that could work ideate for me once again it's going to go ahead and scrape the entire subscriber database all of the different ideas that are working and then propose me a bunch more video ideas that way i can get to 100 200 300 ideas just to click of a couple buttons i'm able to generate all of these ideas super duper quickly because it is assisted by the subscriber mcp which previously would be very very difficult and it's not like all of these ideas would be a hit a lot of people will just kind of prompt prompt AI to magically give them the one winning video idea that's going to take their channel to the moon.
What you actually want to do is you want to generate 100 ideas, filter it down to what are the top 20 ideas, filter it down to what's the top five ideas, and then pick the one idea that you have the most confident in that you can go ahead and create next. Over here, write seven script templates that print views in a niche, why your scripts are boring, and the one structure that fixes it.
So the storytelling people watch to the end. I studied 100 channels that blew up in 2025. Here's the pattern.
That one's actually not bad because I studied it. is really performing very, very well right now. How much a 100 ,000 subscriber faceless channel actually makes.
That's actually a pretty, pretty good idea. The faceless is just quietly printing views. I built a faceless YouTube channel using only AI.
The AI workflow I used to write 10 scripts a day. How to clone your voice so AI scripts sound exactly like you. How I grew subscriber to 100 ,000 MRR using nothing but YouTube scripts.
That's actually pretty good too. So you can see over here, you're just mass producing these ideas. Go ahead and ask them to generate more ideas.
One of the ideas that I actually really like is I really like. this idea so i'm gonna go ahead hey give me 10 title variations for this based on outliers bang it goes ahead and searches our database for all the different ideas on like okay youtube channels with ai and then this is right we built a faceless youtube channel using only ai here's what happened right makers the single biggest outlier video how i built a faceless youtube channel using free ai tool start to finish free tools is actually a pretty big vector i built a faceless youtube channel with ai that makes money while i sleep i built an ai agent to run an entire faceless youtube channel no code all these variations are like very good i actually really like this i built an ai agent to run an entire faceless YouTube channel, no code.
I use only AI to build a faceless YouTube channel from scratch. I built a viral faceless channel using only AI. All of these are very usable and it's kind of about narrowing down our focus.
Just like that, I think it's very, very clear how you can generate viral ideas like this. And you can see all of it is very data -backed. It's exactly what's working on YouTube, which is exactly what you want.
You don't want to use Claude to just go ahead and generate random ideas that you have no confidence in. You want something that you have a high degree of confidence. in that way you can go ahead and generate more and more video ideas just that way so now youtube scripts previously we talked more on like the idea side of things now we're going to talk about exactly how to execute on your video and none of this matters if what you say isn't interesting if how you think about how to execute the video isn't interesting at all i'm going to break down how to use claw to write the youtube script for you so on this channel there are a bunch of tutorials on how to write youtube scripts i have a bunch of youtube script writing courses you can go ahead and watch those but here this is going to be the 80 20 about how to use ai in order to write your script for you in terms of a youtube script there are a couple of components that really matters number one it's your topic what you are actually talking about for this specific video it's clawed full course to grow youtube that is the topic that is being talked about and then there is the title title will determine what is the angle of the video how you are talking about something because it's very rarely just what you talk about it's how you are going to talk about
something, how you are going to make your video different from everyone else's. If you can really nail down what the angle is of your video, I think that is like 90 % of the script. Even if the script is not so really well written, but if you know how you're differentiating yourself in terms of what you are saying, it's going to make the script writing just so, so much easier.
And then next we have the template. Pretty much whenever you are writing a youtube video or creating a youtube video you can boil it down into eight different core templates so in terms of the core templates it is a challenge style video so this would be a mr beast type video that would be considered to be a challenge style it's like hey i'd be able to accomplish this that is basically a challenge style video a documentary video is what many of you guys would have probably watched it is documenting person an event right and it's more of like a third person perspective where it's like documenting it and then you maybe can inject some of your own personal perspective to that there's a listicle listicle is like a top 10 or i'm going to tell you the three reasons why you should start a youtube channel that would consider to be a listicle to a storytelling video so this could be a fictional storytelling video you're not necessarily giving any specific education but you are telling a specific story it could be a fictional story could be a real story and you can kind of see there is some overlap here between some of these templates so we're going to talk about it a little bit as you can see on the screen or maybe there's like an educational video something like this where it's like a course where you are learning something
where the outcome of watching the piece of content is that you learn something from the piece of content. You can kind of see over here, even just in this one, two, three, four, five different templates that I've given you, there are news style videos too. There are commentary, there are review style videos too.
So those are other templates, but you can kind of see that there is some overlap, right? Not... any single video will lie within just one kind of style for example mr b style could be a challenge style plus a storytelling video it could be a mix of challenge style plus storytelling so there is kind of a mix there or a documentary doesn't necessarily just have to be you're documenting something it could be an educational documentary where the goal of the documentary is that you teach something to the viewer it could even be a storytelling documentary where the documentary is aimed at entertaining but has kind of an interesting story behind it that you really, really want to tell.
You can kind of see over here that it's not really black or white, the type of template or the type of video that you're making. And in fact, storytelling documentary can shift in terms of its percentage. Maybe it's 70 % storytelling and then 30 % documentary, where it's more focused on the story and not so much on the documenting.
Or it could actually be 80 % documentary and then 20 % storytelling. There are storytelling elements baked into the documentary. Kind of like an ebb and flow in terms of the percentage.
Every single video... is a little bit different. Even Mr.
B's videos sometimes could be 90 % challenge and a 10 % storytelling. And then more recently, Mr. B's videos are more like 90 % storytelling, 10 % challenge style.
The examples I have over here, educational could be 30%, documentary could be 70%, and that could lead to a business documentary. Or a documentary could be a 20%, a listicle could be 80%, and that can be a documentary listicle. Or it could be a 50 % educational video, 50 % listicle, and that would be an educational listicle.
It's not as important. to nail down exactly what is the percentage split between these two. A lot of people will get very, very granular as to, oh, it has to be 30 % documentary, 70 % storytelling.
And so I have to really nail it. But really it could be 40, 60, could be 50, 50. It's important to just have this in the back of your mind whenever you're really reviewing your script or when you are prompting Claude to tell it like, hey, it should be written this specific way, taking into account roughly these percentages.
Do you have to nail the exact? number in terms of these percentages. No, not really.
You don't necessarily have to, but is it good to have it in mind so that you can guide the AI exactly how to write the script? Yes, that is a must. That is your templates here.
It is ideal to know what type of video that you are making, which will actually drive how you write these templates. And then next you have your brand voice and tone. When you are scripting or creating a video, there has to be a certain tone, which will guide the feel of the video.
And you want to make sure that it is on brand. If I am just reading off a script right now and it doesn't sound like me and it's written by AI, then there's no point of writing it with AI because it actually dilutes your brand.
It makes you stand out way less. You need to really nail down the brand voice and the tone of the overall channel before you can go ahead and write your script. Understand your topic, what your title is, what your angle is, your template, your brand voice and your tone.
You can actually write any script with these understandings. You pretty much have to define these terms before you go ahead and write a YouTube script. There's the research portion of it.
which we will talk about in a little bit, how exactly you can do it, and what a bunch of these new different tools you can go ahead and use can allow you to write this YouTube script. The old way of using AI to write scripts is one problem to research, analyze the outlier reference, which will... allow you to create your template.
And it's like, hey, copy the research of this guy and then write the script. But the problem that you run into is that it actually creates a bunch of AI writing within this because you just have two prompts. Basically, information isn't as direct.
There are a bunch of hallucinations because the research isn't thorough. So the AI starts making up its own content. This way of writing scripts is how 99 % of small YouTubers write their scripts.
To me, that is really, really terrible because this really dilutes your brand. It's really dilute your voice, even on a faceless. channel level this is unacceptable because this is what all of your competitors are doing if all of your competitors are doing who think they can get a way of just putting some research together analyzing some outliers and writing the script you're not going to stand out from all of them on a script writing level or on a content level a lot of script writing is essentially finding ways that you can stand out from your competition if your content sounds exactly like theirs there's just no way that you can win you need to be better or you need to be different so i'm going to show you exactly how i think about you know the script writing process once you have defined all the terms that we talked about and what is the different processes that you can create agentically you would actually go ahead and do in terms of a research there are a bunch of different tools right now that you can use the two main tools that i use is either perplexity or google deep research api the perplexity deep research api is one of the best in terms of research it gives you really really good in -depth data google deep research api is also really strong in terms of giving you very very
depth data. Why do we use the deep research API? API is basically an endpoint, which allows you to use Publixy inside of cloud.
It will only give you five sources, but majority of the times, whenever you are doing deep research into a specific topic, you want to have many, many sources. And that's where the deep research API gives you 50 to 60 different sources, which is exactly what you want. When you have this deep research API and you get all of these different sources, what you then want to do is then you want to verify the research.
You will have a prompt that goes hidden researchers Based on all of these things that you have created over here, tell it to go ahead and research my topic.
It will go ahead and create a research report. You'll get it to verify the research. It will compile the research report to make sure that the research report is good.
So not only do we use the research APIs in order to put the research together, but we also verify the research to make sure there is no misinformation there. There's no content gaps within there. And then once you verify the research, you want to fill in any gaps that you have within the content.
And once you verify the research, you compile the research. report so now you have a baseline for all the things that you have to do and by the way this is the exact human process that you would have done before ai go out there research research reddit research twitter research all these different forums to get information on the topic and then you want to verify the research to make sure you have no fake news and then you compile a research report a lot of what we're doing here with flawed or with ai is essentially taking a process that you would normally do in a human being and automating that what many people have done is that they've just skipped steps that a huge would normally do.
What we're trying to do here is mimic exactly what a human would do. So to minimize amount of error there is so that you can have it as a complete research report where typically you would have research as one step. We don't just do research.
We do deep research and then we verify the research and then we compile the research report to know what we can say and what we cannot say. And only then we don't actually go ahead and write the script immediately. We take all of this and then we write an outline of the video and the outline of the video will be determined by the template then you can see over here this template it'll be the outline right what we say in the hook what we say in the body open loop payoff open loop payoff open loop payoff and then the conclusion in terms of the outline i really like using gpt 5 .5 in order to write this outline more so than claude if i'm being honest one of the main benefits of using gpt is that gpt is much more intelligent in terms of logic and reasoning and because it's much more better at logic and reasoning outlining is basically okay taking a research report and then re where part one should be, where part two should be, where part three should be inside of the script and what should be included inside all of that, much better at reasoning.
And it's not as much as the cost of GPT 5 .6 Sol. I really like using GPT 5 .5 for outlining the script. And once we got the outline of the script, then we can go ahead and write the script because then we have given a guidelines on exactly what to write.
So then over here, we would use Claude Sonnet in order to write it. We have also split up how we write the hook. We don't just write one hook because as you and I both know, The hook is the most important part of a YouTube video.
If you fail to hook the viewer, viewer will leave and they won't watch the rest of the video. So it is vastly, vastly important to really nail the hook and to actually nail the hook, feed the AI enough data as to what a good hook is. What's really useful is if you can feed YouTube videos on how to write a great hook, that'll be really useful for it to know how to refine the hook.
And more so than anything, if you can feed it 100, 200 best hooks that you have written, get it to. refer to these whenever it's writing hooks, then it will have a lot more data, a vast amount of data set that it can use in order to go ahead and write the hook. What you want to do when you're writing hook is you don't just want to write one, you want to write three to four different hook options based on different frameworks and then rank it.
and then pick the best one of the hooks in order to go ahead and create the video with. We actually spend a lot more time on this hook just like a normal human being would. When you are writing scripts at the very, very top levels where we have written scripts for 100K, 200K, million subscriber educational YouTubers, we write three to five different hook variations based on which one we think will perform.
And then inside of YouTube Studio, we can go ahead and edit it out in post. You want to make sure that... you have many, many hook variations just as you would as a human.
And then you pick the best one by giving it all of the different data sets that you have inside of your head or that a master script writer would actually have. So that's mechanically how we think about the hook before we present to you the best hook option. And then the next step to go ahead and remove repetition.
Now it's important to remove repetition because when you go ahead and write scripts with AI, one of the things that AI does is that it actually repeats itself multiple times. It would say it's really important to do this it's really important it is really positive so it will have three to four sentences really reiterating and explaining the same point over and over again but that is not how YouTube works fundamentally that is not ideally how you want to execute a YouTube video because with any given YouTube video what you really want is to explain a point fully in the least number of words possible but as long as you need and then cut off all of the repetition that is what good writing and a good piece of video looks like where you don't have to over explain something and that's where removing repetition becomes really really key you want to remove all the repetition so we get the ai to go ahead and look at the script see what we can cut from the script and then it goes ahead moves it to the next step
So removing repetition is that step. And then this step over here is to humanize it because we all know that AI has these like AI writing tropes, right? Inside of the brand voice, we can tell it to not use certain words or not use certain ways of phrasing things, but it often ends up missing some of these.
You want to remove all of these. ai writing tropes one of the pro tips that i can give you when you're using this humanized prompt that i'm going to give you in the link in the description is that whenever you are humanizing any piece of writing you don't want to humanize a full thousand piece of word of writing What you want to do is you want to humanize it in chunks.
So let's say you have a whole script that is 4 ,000 words. You want to humanize it 500 words, 500 words, 500 words, 500 words, 500 words, 500 words. You want to chunk it up because once again, you're basically telling it to rewrite the entire script.
Here in this step, we are just removing parts. We're just canceling parts that are too additive. Over here, we are getting it to rewrite the entire script.
And just like how we wouldn't rewrite the entire script without... an outline you don't want it to rewrite the entire script but basically you wanted to chunk it together until 500 words humanizes humanizes humanizes humanizes this is the sauce over here you want to humanize it in chunks and that way it can minimize the amount of ai writing because sometimes it just forgets right no matter how good the model is sometimes it forgets and it misses it especially if your prompt is very very long sometimes it just completely misses it where hopefully one day it will not once you have humanized it You'll want to do final voice checks and refinement.
The reason we want to do final voice checks is because actually when you go ahead and humanize a script, you might lose some of the voice because you are trying to swap certain words for human words. But some of these human words aren't even words that you would actually say. And because some of these words aren't words that you actually say, you will then have to cut it out again.
So then you'll want to do final voice checks because you have your initial voice profile. You'll want to remove all of that and remove all your writing and actually refine a ton. of that so that's why we have final voice checks and refinement in terms of the research side of things all of these steps where you go research verify the research compile the research report outlining is a single step and we can only do outlining as a single step because we have really compiled this research report and done such deep research there when you write the script it's not one prompt where you go ahead and just write the script and call it a day you have to do five different steps before you can get like a full good youtube script and this is the most simplest part of it this is the workflow that you'll want to have there is really no other process to replace this no matter how good the models come up with this is exactly how i really think about it previously you would
have to you know prompt each of these one by one until you get like a good final script but thank god you don't need to do that anymore because you can actually run this as an agentic workflow you can create all of this inside of claude i'm going to show you a simple way go ahead and write these scripts using this tool that we have subscriber pretty much this script was fully written the process that i tell you about you can see that it's using claude's on a five it's for one of my clients fiction profits academy they do a couple million dollars every single month with their youtube channel over here if you would look at the details right you can kind of see all of the different files that led to this, you know, being written, all of the different tests that we run it through, all of the different skills that we apply, how long it takes in order to write the script.
So you can see over here, it takes 1 ,800 seconds. So this took 30 minutes to actually write. When you're going to just use Claude in order to write the script, that will take you maybe...
five minutes to put together a script with kind of this old way of writing the scripts which is ideally not the way that you want to do it because it is so simplified versus you're competing against someone like me who is running 30 minute scripts right through a multi -step agentic approach in order to get scripts better than you you have to write scripts that are equally as strong and actually this is still using Claude because we have Claude Sonnet integrated inside of this we can see we have all of these different models for writing over here Claude Sonnet 5 being my favorite right so we actually use Claude when in this and what we're going to do in terms of subscriber is one of my favorite script writing tools it's because i'm one of the owners of it i use this every single day what you're going to want to do is that you're going to want to connect your cloud subscription onto subscriber and by the way you can get started with subscriber for free today check out the link in the description if you go into your team settings over here if you click on agent mode what you can do is you can just connect your cloud subscription previously we had it where you needed to connect your api key to subscriber in order to write these scripts now no longer you can just connect your subscription to it and
and it will just use your subscription limits. So if you have a $20 subscription, $100 subscription, $200 subscription, or if you have a ChatGPT subscription, pretty much everyone, if you're watching this, you probably have a ChatGPT subscription. You can go ahead and use that in order to write your script too, and you can use those models inside of Subscriber.
If you were to just go ahead and write the scripts, all that you have to do in order to start this process is you can just start a new chat. over here you can connect to your youtube channel and then once you go ahead and click create script you just have to tell it what topic that you wanted to write for your channel okay so i'm going to show you how we got the final product for fiction profits academy so create a script about how to write amazon books so good they sell themselves then you want to tell it you literally just follow the prompts here structure it like an educational video focus on how to write amazing content for amazon and bang you literally just follow this entire process in terms of the research steps over here you can kind of see do have like perplexity google deep research api in order to find the sources or what would be really good is if you can include kind of your own thinking and your own methodology inside of this so it will actually ask you a bunch of questions here you want to write a script about writing books so good let me come from a few things right here's your length here's your angle let's just say yeah this angle works 3 000 words work too
let's write the script basically what you want to have within this is you want it to have specific uni ago so what it does it goes ahead and it goes and does research on what is potentially the best angle that you can use inside of your video because subscriber is trained on 40 million pieces of youtube content data it can go ahead and you'll scour the internet all these things and you can kind of see over here it just opens a canvas cloud interface within this and you can go ahead and you know write this educational template you can select the template choice over here you can see all of these different models that you have i like using clouds on a five of course especially because this is a claw course i like to use this educational template because we already have a bunch of these built and templates within here which kind of explains how to use each of these in this case it's an educational video right because we are teaching something the voice it just goes ahead it takes videos from your channel and then it learns the voice so it automates a lot of this process that i just showed you where it helps you figure out the topic the angle it has a bunch of preset templates and then it helps you figure out the brand voice so this is all that you need to write a script we'll go ahead and write some of these over here it gives you a bunch of questions and for you to answer so you can go ahead and add research you want to add your website you can go
FictionProfitsAcademy .com. Add your website over here. And then what it will do is it will process the URL and fetch the content.
So it has a bunch more context on exactly what you have here. And then also you can go ahead and add text. If you have libraries of text that you want to add, you can go search YouTube for other videos.
You can add that to your research too. Or you can tell it to go do AI research, which is exactly what it will do. You can answer a bunch of questions.
And what's really nice about this, literally just talk directly to this. So you can tell it, hey, like what's specific writing? techniques or tropes that you personally found most effective for keeping readers engaged in terms of you know what writing techniques have really worked it's like having open loops having really clear titles attention -grabbing titles that tell the viewer exactly what they're going to get and inside of the book instead of in each chapter giving the viewer directly what information that they'll get from reading the chapter you want to start with a story to really sell them on the importance of why that specific chapter is so important every single of the greatest books always starts their chapter with either a personal story or an anecdote of a fictional character.
It gives the reader why it's important, not just exactly how to execute on it or like why they should do it in a certain way. Can you share a concrete example of a book that sold itself through quality content alone? For example, Alex Homozy's book, $100 million offers, you know, sold itself.
through the way that it was being written and itself a really really good offer like ten dollar book and the content was really really good and he pretty much did no marketing for it and he sold a gazillion copies like many many copies there and then you can do like voice dictation all of that and then you can answer these questions and i'm just gonna answer these questions anyhow this is just demo video you click continue script and you can see over here it pops up the script canvas or the agent mode so the agent mode is the one that will take you through this entire process i've already explained it where you can just tell it to do public deep deep research yes you can tell it the cues that you want you can tell it the model that you want i'm going to use on five and then you just click bang run agent mode and then basically it's just going to start writing the script within this and if you give it a couple of minutes this should pop up over here where you can kind of see the entire script is being written and it's pretty much ready to record this is the exact same methodology that you can go ahead and write educational scripts ai written scripts i believe this is the easiest most efficient way in order to write really really banger scripts that is in your voice and it's all this is written inside of claude that is it
in terms of the script writing portion of this video. So now that we're done with how to write a YouTube script, we're going to get into exactly how to create a banger thumbnail before we can go into how we can use AI in order to create a banger thumbnail. We have to understand some of the core underlying principles of what really makes a great thumbnail from a theory standpoint.
The next 40 minutes will be dedicated towards giving you a full understanding of what makes a great thumbnail because a thumbnail is like 70 % of your video because if you don't get the click, no one is going to watch your video. So you need to be able to get the click from the viewer.
That is one of the most important things in all of YouTube. And I have this dedicated assignment teaching you exactly how to create a thumbnail, how to do thumbnail research. Let's get right into it.
now in terms of creating thumbnails with ai i'm going to show you how you can use subscriber for it and also i'll give you a couple other tools that you can use let's say you don't want to use subscriber for it we have this channel over here you can see like these videos are just absolutely crushing 155k views and you can see over here what it actually has is it has a repeatable format of a thumbnail really really works well because when one thumbnail works well you basically double triple down on it and basically that's what he's done this entire channel which is really really cool to see when you can kind of see that there are a few very variations here, the economics of owning a gas station.
So you want to buy a charge station. So you want to buy a pharmacy and we'll show you how to create these thumbnails inside of subscriber. Now over here, I just have an overhead channel written over here.
You can see there are a couple of scripts that are being written for this specific channel. If we go into the thumbnails tab over here, you click into thumbnails, because there are really two ways in order to generate these thumbnails. One way is to use a reference thumbnail to go ahead and create these thumbnails right here.
So you can see over here, the video title, let's say we're going to use, okay, the economics of owning. a supermarket and then you tell it what's the video about you don't really need to give context you just tell it that and then you can give it a style reference so let's say you want this i'm just going to pick a random video this one has 155 000 views so i'm just going to go ahead and paste the style reference one of the things with ai is that it's really useful If you can directly give it exactly what kind of style that you're looking for.
If you can tell it exactly what style that you're looking for, it's going to have a really, really good time in terms of replicating that type of thumbnail that you are going to go ahead and generate. What this has done is that we've given it the reference. It knows exactly what the title is.
And because it knows what the title is, they can go ahead and then create a prompt that you can then go ahead and feed inside image model. In this case, it's subscribe. image model really digging deep into what the specific style is identifying the visual elements so they can go ahead and create the thumbnails literally creating thumbnails of ai is one of those i think magical things because previously where it will take ai many many hours in order to generate you know a good image now it can copy images like one to one or it can generate certain concepts from scratch we'll kind of see how this comes out in a little bit so this is the only guide that you need to master the craft of YouTube thumbnails.
So here are just some of what you'll learn today. So today we're going to talk about the psychology behind the click. So it's understanding the curiosity gap that makes viewers unable to resist clicking.
The psychology is everything. We're going to go over 10 plus different proven thumbnail frameworks. And these are frameworks that you can just copy paste to immediately improve your click rate that you don't really have to worry too much about.
We're going to also talk about advanced principles, the 10 advanced principles, how the top 1 % engineer their thumbnails to make them consistently outperform. We're going to talk about AI -powered thumbnail creation, so how to create professional thumbnails in minutes. We're going to be talking about A -B testing mastery, so how to know exactly when to test, what to test, and how to interpret those results.
And then lastly, we're going to talk about fatal mistakes to avoid in every aspect of the process when you're making thumbnails so that you can stop making these errors that are tanking your click -through rate before you even upload. And we're going to go over so much more. So stay tuned.
The thumbnail is the very first thing that viewers see along with the title, right? It's what they use to decide whether to click or not. And it's the hinge where all else depends on.
And your thumbnail needs to stand out. Because viewers scroll passively until something visual catches their eyes. So your thumbnail needs to stand out.
Because when you scroll on the YouTube homepage, you are presented with all of these different options of videos to watch. And over here in the sample example, we have eight videos that is presented to the viewer and there's only one winner. Meaning that YouTube is a zero -sum game.
When someone decides to watch one video, all the other videos loses out. And what kind of dictates what gets the click? Majority of it is a thumbnail, right?
And it's a title, right? It's 50 -50 between the thumbnail and the title. And this is super duper important to understand because it dictates how much effort you put into everything, right?
Because if your video is like the best video in the world, but the title and thumbnail doesn't stand out, doesn't get the click, it loses out to the seven other videos that people, that it's being presented, you get absolutely cooked there. Okay, so the thumbnail sells them on looking at the title, the title sells them on the click, which sells them on the hook of the video, which sells the intro and then the rest of the video.
And it's, the thumbnail is integral to the success on YouTube because it's actually what stops the scroll, okay? Because... When the viewer goes down a certain journey, they go from the thumbnail to title, title to click.
So that means a thumbnail comes even before the title. So you want to have an optimized thumbnail. So a good thumbnail leads to a bunch of clicks and a bunch of people watching your video, and it really justifies you putting a bunch of time into your video, and a bad thumbnail still clicks.
And the majority of people have bad thumbnails, so actually you're in luck. Once you understand a lot of the principles that we're gonna talk about today, you'll be ahead of 99 % of people. but just getting the click isn't enough because I can very easily click bait you with, you know, a cat video.
But the goal of the thumbnail and the title is to capture high quality attention and as much of it as possible because it's the first point, first touch point of the viewer, even before the title, right? So remember that not all attention is equal. Because you don't just want to attract any random viewers, you want to attract your ideal customers or your ideal viewers because they are people who can give you money or will help you monetize with AdSense, help you monetize with these sponsorships instead of a bunch of, you know, little Timmy's who watch my current videos.
So if you were to optimize your thumbnail, you don't just want to optimize for the click, you want to optimize for getting the right person to click on your video. And if you want to do all of that, you need to nail the psychology behind the thumbnail. Right because the psychology behind the click is in essence a curiosity gap Right and a curiosity gap is the gap between what we know what we currently know and what we want to know there's a big gap there and Really, not all attention is equal right because you want to attract your ideal customers so that they can buy from you right and Understanding is this is super important because there are two critical components here with your thumbnail Number one, it's there's a killer curiosity gap and number two is that there's a killer scroll stopper So once again when people are scrolling on YouTube, they're mindlessly scrolling scrolling scrolling figuring out what video to watch the thumbnail must stop to scroll and Then once people look at the thumbnail, there must be some kind of curiosity that gets them to click So stop the scroll and I get into click Okay, these are two different things and your thumbnails must achieve both if your thumbnail only achieves one
If there's a curiosity gap, but when people scroll, they don't see it, that means the colors don't really pop, there's not enough contrast with what's existing out there on YouTube, and you get absolutely cooked. So the five categories of thumbnails that use a curiosity gap are, number one is a moment thumbnail, okay? So a moment thumbnail drops up.
drops you know the viewer into the moment before a powerful reaction so you can see over here both of them are in a private jet this kind of sets the scene and both of them are talking right i quit my nine to five then i made a billion dollars what does this signal right this signals that hey they're in a private jet so it's probably someone pretty rich and then you look at the title asking a billionaire so it's like okay that makes sense so it's super rich private jet i quit my nine to five and that made a billion dollars so there is curiosity baked into that right so noah kagan I wrote a brief description over here.
No kicking the guy on the right. He's leaning in and listening attentively. So you can see that it's in the moment.
It's an emotion, right? And the billionaire on the left is in a hoodie and sneaker. So there's kind of like a contrast over there.
And then there's the quote, which invokes the curiosity. So dropping people into the moment. and a lot of this can be done by you know when you are writing something when you're working on your laptop and then you use that as the thumbnail so you can also really use that because that illustrates that point of hey it is in the moment right it's in the motion over here a story thumbnail so a story thumbnail introduces a point of tension or question that sets up the first act of the story so you can see here on the top right This is Jordan Walsh thumbnail, right?
Minus 30K to $6 million, my story. It's got, you know, two plus millions, 2 .5 million views. So this illustrates a certain story, right?
This illustrates a transformation. So this is a transformation plus a story thumbnail. The video is on, like, my story.
So you can see over here, it introduces, you know, a question in the viewer's head, right? It introduces contrasting colors. And it depicts people like, hey, there's a story that's to be told.
So for example, in this Eric video, right? This Eric thumbnail is a novel visual. So there's like some sort of novelty there.
And then there's kind of a story, right? Sneaking into a secret VIP club, right? So it invokes a bunch of questions, invokes a bunch of curiosity with the story.
What will happen in the club? What is it like in there? Why is it so bright and exciting inside the VIP clubs?
Which makes the viewer want to find out more and hence they click. So that is a story thumbnail. So the thumbnail tells a story and invokes a curiosity gap.
And then there's a result thumbnail. A result thumbnail depicts a dream outcome of the viewer. So it clearly showcases the idea of the result that viewers aspire to, and this could be a business achieving explosive growth, a person losing weight, et cetera.
And the emotion you want to spark is, hey, I want that too. So for example, over here, this is an instantly AI thumbnail, and this very clearly shows this is a Stripe dashboard, $10 ,000 in the background, monthly revenue, and then the graph is going up. Right.
So it's meant to mimic a Stripe dashboard. And this is something that most businesses are familiar with. So not only do we use kind of an asset.
Right. We could have used a green graph here. We could have used like a blue graph.
Right. That you might see and see in a couple like stock apps. But this is specifically a purple graph because this is something that the viewers and the business owners.
So I built a secret business. to prove it's not luck and so it's something that business owners would recognize so he wants to outline the dream outcome inside this thumbnail and you see it says i built a secret business so it's him with his finger on his mouth so you can see that over there and then there's a transformation thumbnail right a transformation thumbnail features a desired result and the beginning state so it clearly showcases contrast between an unhappy frustrated before state versus an impressive after state.
So there's usually a visual cue, like a split image or arrows to emphasize drastic change from one state to another. Right, and the emotion that you spark here is curiosity. So once again, this is a transformation thumbnail, kind of like the Jordan Walsh thumbnail that I showed you before, but you can see over here, left is 1K, right is 8 million.
So I went from 1 ,000 to 8 million, my story. so it is another my story video but inside the thumbnail it's a transformation right so transformation is really great for illustrating a result and a story so you can see here if we were to you know dig a little deeper into the thumbnail and you were to look closely instead of just looking at the surface level right what most people will do is that they'll look at the surface level and then they'll click on the video but if you would analyze the thumbnail you can see on the left right the left half of his face or i guess his right half of the face tire in pain broke and then there's 1k right darker color scheme and in the after shot his expression is now joyful showing strength confidence of a lamborghini and eight million dollars right so there is stark contrast here And then there's our novelty thumbnail, right?
The novelty thumbnail draws power from our natural urge to experience something that we've never seen before. The video idea should feature something incredibly strange, unique, or unexpected. The more something deviates from their daily life experiences, the more curiosity it will spark when featured prominently in the thumbnail.
So this emphasizes sensations, showcase something that looks like it will feel interesting to touch, move around, interactive, and the curiosity will be sensory as well as visual. So this is the mr. Beast example So this is mr.
Beast thumbnail of a real -life laser maze is something that you know, we've really never seen before So there's a novelty aspect to it, right and you could see if you look at the details of it. He has a shredded shirt he's lying down his face is like ah over here and The lasers look like it's about to kill him and you know on his face.
There's kind of a scratch So it shows danger within that so I write a brief description to kind of describe each thumbnail So you can see over here that there is a level of attention to detail that goes into the thumbnails here Right if this thumbnail would be made and his shirt would be like a proper shirt It wouldn't be realistic, but you can see there's all the different hairs inside a shirt If you want to grab people's attention you need to master pattern interruption your thumbnail has to make the viewer stop scrolling And pattern interrupt elements you can use that are proven to make people stop scrolling is like faces, large numbers in the background, some kind of money, danger, emotion, you know, bright colors, or being aesthetically pleasing, right?
All of these leads to the click.
Now, I want to talk about the three C's of thumbnail design, right? And once again, this is going to be a comprehensive guide on everything thumbnails. Let's get into it.
Okay, content, composition, contrast.
So the content is any visual elements, right? This is the people, icons, graphics, background text, and whatever you need to drive curiosity. So the content over here, everything that I circled is basically all of the different icons here.
This is all considered content. So before you design a thumbnail, you're like, hey, what do I want inside the thumbnail? So you think about what is the content of the thumbnail?
and then there's composition right it's symmetrical like main character in the center left and right are similar or asymmetrical where the main character is on the left or the third grid you can see here i have a bunch of examples here that are circled so you can see this one on the top right this is evil money and then there's like the jesus like thingy and then uh there's michelle kari over here she's on the third so there's kind of asymmetry here and then there's contrast right so it guides eyes and captures attention by playing with you know the brightness the saturation or the hue in this specific case right hack number 20.
so you can see over here all of these thumbnails do stand out in some way right and they are scroll stoppers and they are not only scroll stoppers but they also induce some kind of curiosity so it gets the click so let's talk about 15 different winning thumbnail formats where i'm going to speed run through these so that you can literally just copy any of these and be like, hey, what format works for my video, okay?
There you go. Number one, title hidden. This is super simple, right?
Face on one side, simple text on the other side, right? The goal is to establish familiarity with the face of authority, and then the text should complement the title, right? And ideally, you don't repeat the exact same things, and we'll go into that in a little bit.
So this is the simplest title hidden, okay? Number two is the dramatic face, okay? The emotion is the main message, but the title gives it meaning.
So you want to be intentionally vague about what the video is about, and this relies on familiarity to draw attention. So if people know who Ryan Trahan is, then they'll click on this video, and you can see here there's a dramatic face. Not much happening in the thumbnail.
And then A affects B. Okay, show one person or object clearly about to affect or target the other. In this case, it's a podcast, right?
So there's two people over here. The guy on the left is embarrassed.
Conversation question, right? Show two people with a question in the middle. So you can see over here, this is very clearly a street interview.
And then what year is it? So this is a street interview. So it could be a normal interview where you do it in the podcast.
But here, this is a conversation question. Tree panel progress. It shows the same subject changing over time or moving through different stages.
So you can see over here, left to right, 100K, 1 mil, 10 mil. And then it says three stages.
Problem state right using a powerful image. So the viewer quickly gets the problem even without much context. So don't show the solution yet So this uncertainty evokes curiosity.
This is a mock robot thumbnail. It kind of looks like The hospital or the building is about to collapse. So this is introducing a problem And also this is you know leaning into our novelty where it's like hey, you've never really seen this before on YouTube So so there's that And then there's the contrast, right?
Place two states side by side, right? So this could be a before -after transformation type thing, right? Play with the size of the color.
In this case, right, there's a contrast and there's also a novelty element to it because there's a, you know, pink chocolate factory over here. And here's red arrow with the me. Don't do this, right?
Shows an action that's clearly wrong. So it's crossing, you know, blanking. right it plays on the fear of making the same mistake don't do this thumbnail so it's a negative thumbnail so negative connotation negative things tends to work pretty good eliminators right show the main rule of the video right set the limitations no money no ads in this case no pull up bar so how is that possible Motion arrow, right?
Movement is implied, but the result is hidden. The arrow could be of a different shape to increase curiosity about what's going to happen. So in this case, Paris time lapse, minus 300 to 2025.
So you can see that there's motion. There's something happening over here and it guides your eye. You know, it guides your eye from negative 300 to 2025.
So your eye moves. And then there's conflict, right? In this case, it's kind of a podcast here, right?
Show two objects interacting with each other. In this case, none of them are looking at the thumbnail and they're like kind of like dogging each other down.
Mild progression, right? So the story is already happening. Show a snapshot of a later stage.
No clear start or end to drive curiosity, right? So this is like day 29. So it shows a mild progression.
So in the title, probably it illustrates like 100 days like stuck in the sky, something like that. Living 100 days in the sky. and you can see that there's day 29 over there and then there's the comment or post show comments in the thumbnail to set up expectations right so you know this can create a sense of conflict or tone right in this case there's the comment or you could use like a tweet type thing that's also considered to be a format now this isn't working on niches but you know like commentary niches like in this case it's movie commentary Accusation, so like this is also kind of the head with the text, right?
But in this case, it's the kind of the feel of the thumbnail. So in this case, the view of you is targeted. You point at or provoke a disagreement and this drives clicks.
And this is a review, right? So this kind of star in the review, nothing crazy here. So let's go into some thumbnail principles, okay?
Because , it's engineering. So let's talk about some relatability, right? So in terms of why people click on something, they click on something because there's curiosity or there's relatability, right?
So relatability can be in the form of a couple of things. Relatability can be in the form of, hey, I relate very much to the person. So the face on screen.
So that means that's familiarity of the face on screen that creates relatability.
uh dynamism right so the best way to build a skill right so using mid action shots capturing engaging moments right so there's some form of dynamism over here this is super duper important this is the rule of three okay and the rule of three states that hey every thumbnail should have three main elements here so in this case it's uh will number six over here he's in the blue jersey the guy on the right number seven and he's dunking and then you have the text with the arrow age 16.
right so there are three main elements over here and it's important to have only three elements because when people look at your thumbnail uh you don't want it to be super duper cluttered right and this really removes all the clutter anything that doesn't get the click it makes every moment uh uh element eligible on mobile right and it focuses on the core message over here so it's like oh like age 16 year old guy dunking on will that's it transformations this one we talked about right so clear contrast in the transformations uh number five is realism right so using genuine unfiltered imagery so in this case the guy on the left over here you know it's not super hyper edited uh that's also kind of in the line in line with um the style of video that he's going for so this is also contingent on the style of video that you're creating number six any facial expressions okay so it's you know showing genuine relatable emotions so if it's a more serious video then you use a serious face if it's shocked right then in this case you use a shocked face like this right and this will allow you to capture authentic reactions express clear emotional states and match the emotion with the video scene so it depends on the state of the video that will determine the type of face that you use and then there's congruence okay your title and thumbnail so you want to make sure that there's perfect alignment between the title and thumbnail
so that you can create a cohesive storytelling, can avoid misleading disconnects, and you want to make sure that, hey, these two guys work together.
Number eight is a visual hierarchy. So whenever you look at any given thumbnail, on first glance, your eyes will shift to one thing. Very rarely is it like...
you know like when you present a thumbnail to 10 different people the first thing that they see is completely different right in terms of what they see in the first like 0 .2 seconds it'll be the same thing right there's a certain visual hierarchy over here so if you were to close your eyes and you open you look at this thumbnail the first thing that you'd see is the woman who is stretching over here and the next thing that you would see is the combat mobility workout the breakdown so now depending on the composition of your thumbnail that will lead to a certain visual hierarchy right so you wanna create clear focal points.
In this case, you know, the woman is occupying more than half of the thumbnail, like three quarters of the thumbnail. So that's the very first thing that you'll see. And that is very useful because she's in an unusual position.
She's like kind of like stretching. It looks like she's kind of kicking. So that is actually to the benefit, right?
And then the next thing that you'll see is the text on the right, right? Combat, mobility, workout, the breakdown, right? Probably could have simplified the text over there, but there is very clear focal point here.
right and you know she's using size and placement strategy to her advantage right if she was you know kind of in the corner of the thumbnail maybe if she was only like a quarter on a quarter of the thumbnail then this thumbnail would completely look different right but in this case she used she uses composition to advantage to guide the viewer's eyes and make the most important element pop number nine is comparisons right presenting contrasting viewpoints right in this case this is a tier list type video so There's that.
Number 10 is the promise. So it's to clearly communicate the value, show the benefit outcome, focus on what the viewer gets, and make the promise super obvious. There you go.
So let's talk about some practical execution, exactly how to design the thumbnail. One thing that I would do right now, if you're watching this video, you likely have access to subscriber AI, right? And you are now able to generate full thumbnails, right?
You can test different variations and play around with the prompts over here. Obviously, as I'm recording this, our thumbnails will be improved and maybe the UI will look a little bit different, but I'm going to do a little bit of a demo so you can kind of see how it works. Okay, let's go.
okay so we're here into subscriber ai if you go and click onto your channel over here and you go into ideas this is exactly where you know like in this version of it where you'll be able to design your thumbnails okay so let's say you pick this idea you're like 12 scripting mistakes that f small channels 2026 update what you can do is uh you can go over here you can click on this video over here and it will bring you to this bar over here on the right so what i like to do is i like to generate titles until i can get a title that i'm happy with and uh you can see over here right there's an immediate prompt that is given to you right and you can add references so i'm like hey i want to add my face so i'm going to go ahead and add my face over here Okay, let's say I want to use this face.
It's kind of a funny face. But then what I'm going to do is I'm going to add reference over here. And what I'm going to do is it says, hey, keep the settings off and settings on panel structure.
Change the text to script off. Redecline the graph and script on. So this is kind of just to recommend it.
If you're like, hey, this isn't good, let's just generate one thumbnail to kind of see what happens. bang you can kind of see over here this is kind of funny where uh you have the face that's replaced because i gave the face and you know i gave a shocked face so this is what's what it's going to give script off on retention drop 70 plus retention which is exactly like what we say over here right if Let's say you're not too happy with the prompt.
You can click this button, Enhance Prompt. And then over here, what it will do is that it will completely change the concept and it will enhance the concept so that it is more YouTube optimized. So you have a lot of control over here.
Let's just kind of see what it generates. But you have a lot of control over here. This is using the latest models that are best for thumbnail generation.
But you have a lot of control over here. If you're like, hey, I just want to change the face, you can change the face. If you're like, hey.
want to isolate a certain element you can isolate a certain element and you can change it that way it's nothing extremely complex and let's just see what we get and bang you can kind of see right the script concept is like totally different if you're like hey i don't really like the font or you can go ahead and change the font right and also like added hands over here what what i think is really cool about this is like hey you have like a lot of control and you can you know create thumbnails like pretty pretty easy but um what what i think is really cool here is that when you ever you generate a thumbnail you can tell it like hey i want to make this change or if you're not sure what changes to make, we have this thing called AI feedback, where it's trained on all of the best thumbnail best practices.
And they'll tell you exactly, Hey, what should we edit about this specific thumbnail? Right. And of course, if you use a bigger picture, the thumbnail will look completely different, but then it'll give you a couple of recommendations.
And then you'll be like, Hey, what recommendations do I want to change? So remove all small access labels. So yes, I would agree.
They create visual clutter. Okay. So it's telling me remove that.
So I'm going to take that. Simplify the text to just off or on. on top of the budgets the word script is redundant given the video title it takes a valuable space so let's just take this over here right increase the drop drop off the side size drop plus thirty percent retention by thirty percent doing to ensure that they had a primary focal point eh not really crop title on the creator's head on shoulders remove the dead space at the bottom and make not really replace the dollar signs on the right with a simple views label no okay so let's take these two pieces of feedback and in here i'm going to go ahead and generate the new version so let's see it follow those specific instructions so basically you can go you know infinite rounds of this ai feedback ai feedback ai feedback until you get something that works very well bang you can see over here it you know it removed the graphs and it added this over here which i think is pretty cool so it literally like you can just prompt it to like
infinity until like um you're good with the thumbnail or you can export a thumbnail you can download the thumbnail once you are good of it and um this works for like face on this works for faceless for example something like this if you're like hey i just wanted to um change the face of it right what once again what you can do you upload references once again you can upload references over here and then uh keep the blue grid in background and the red box change the subject to the creator looking intensely at a printed screen there you go like let's just click here generate thumbnail see what it comes up with and bang you can see over here this is a button it actually manipulates the face based on the prompt or you can just change the face one to one so i think this is like pretty cool and yeah that's how you can use subscriber in order to create these thumbnails Cool.
Let's go into some of these concepts towards the end. And we want to talk about the guidelines for taking your thumbnail photos. So.
The difference in quality is massive whenever you are taking thumbnail photos, okay? So if you do a screenshot from like a video recording, it's lower resolution, it's inconsistent lighting, it's unflattering angles, it's limited expression control. Whereas if you do a dedicated photo shoot, actually this dedicated photo shoot doesn't even take that long.
It's like, hey, finding a wall and wearing a white shirt in order to take these photos. But if you use like an iPhone, which is what we recommend, you get 4K quality, you get professional lighting, you get intentional framing, full range of emotions, right? You can get all of these emotions.
you can literally get it with one photo shoot. So these are 11 essential expressions that you need in your photo library, okay? So we have smiling over here, self -explanatory.
We're smiling, open mouth with teeth, laughing, right? Having multiple variations of you laughing, right? Angry, straight face, shocked, open mouth, upset, branded, right?
Where you have certain branded expressions over here. um and yeah those are all the different types of expressions that you want to get right uh below this video you have a list of all of the different expressions that you have to get uh and we have that over there okay so just get all these different expressions whenever you're doing photoshoot uh pretty self -explanatory there in terms of photoshoot best practices right you can use an iphone a lot of people ask me like hey do i need to use a professional camera i would say hey an iphone is good enough or honestly like samsung or like all these other phones the camera is honestly probably better than iphone but uh literally just use a iphone that's what i recommend that's what most people have so you want to make sure that you shoot in a 4k resolution and with a 60 fps right to make sure that the photo quality is actually sharp sharp sharp Never use webcam or low -quality recording devices.
This is pretty self -explanatory, right? And you also want to take individual photos, not video screenshots. So if you were to have your phone, you can just take as many pictures as possible or spam it, right?
So that way you can have a higher quality pictures and have multiple pictures to choose from. You want to avoid colored lighting in the background because then this will distort and make it harder for the thumbnail designer slash the AI. You want to look directly into the camera.
right and you want to have the frame from the shoulder up in this case right we we took it in 4k so you can actually zoom in and actually the quality will still be good And whenever you are taking these thumbnails, you want to make sure to sell the emotion, right? You want to take multiple variations for each expression and you want to organize them well in an asset folder.
Honestly, I don't follow this rule too much, but you want to just have it in an asset folder. And you want to exaggerate the emotion as much as possible, right? And a lot of you guys are going to like feel kind of goofy, like taking these pictures.
But trust me, it's essential to take these pictures at least once because you never know when you might need them. And when you need them, then yeah. So a couple of technical guidelines, right?
With the actual design of it, right? This is a text guideline, okay? On the thumbnails, I always say five words max, right?
We always use a sans serif font. We never repeat the title, right? Because the title communicates a certain idea already.
And when the title communicates a certain idea, if you repeat it with inside the thumbnail text, you're kind of wasting space. So we want to make sure that, hey, the title doesn't match the text. The text should be additive, not repetitive.
So don't repeat the title. Optimize it for mobile viewing. And the main character should always dominate over the text.
Right? Once again, five words max. Sans serif font over here.
Okay? Use sans serif font. Like, it's the best.
Kind of see examples over here. Right? Never repeat the title exactly.
Right? This is... On the right here, how to learn so fast it's almost unfair, right?
And on top, it's master anything. So it's additive. You can see eight psychology -backed design ideas for an energetic home, draining versus energetic.
So you can see both over there. So once again, the text should be additive, right? You don't want it to repeat.
You can see over here on the left, easy and affordable teriyaki chicken meal prep, better than Panda Express. Pretty good. How to start a one -person AI business, right?
This shit prints money. Charity prompting techniques no one talks about. and then it's master chat gbt so it's always additive it always adds a different layout so you want to make sure that the thumbnail text doesn't match the title text okay the main character should be you know should dominate the text right over here on you can see like main character is like in the center of it so these are just examples so you can kind of like see these concepts in life optimize it for mobile viewing right uh so i'll show you guys a tool where you can kind of see uh the thumbnails being previewed And we always go through these three simple tests.
It's the clarity test, it's the contrast test, it's the glance test. Clarity test, can you see all elements in smallest YouTube size? Contrast test, does it stand out among competitor thumbnails?
And the glance test, show to someone for two seconds, can they fully understand it? So we have this thumbnail checklist. I'll also link the thumbnail checklist down below.
But a couple of things that you want to go over, is there an identified clear curiosity gap? Right. And if you can't clearly state the curiosity gap in one sentence, then it's certainly not strong enough.
OK, we want to make sure that we choose the appropriate thumbnail type. Right. Whether it's a how to how to whether it's a moment, a story result, transformation, novelty.
Make sure that your video content targets the primary motivation for your target audience's primary motivation for clicking. Optimized text. Right.
Selected three elements minimum. So we want to make sure that it follows the rule of three. uh we want to make sure that it applies the proper composition right so whether it's symmetrical asymmetrical whether it is leading lines and whether there's a clear hierarchy and yeah that's basically the thumbnail checklist so now let's get into some common thumbnail mistakes that are killing your thumbnail okay so we just went over all the things that you need to know when you're creating your thumbnails let's go over some thumbnail mistakes So lacking contrasting colors.
So you can see here on the left, blue and yellow are contrasting, right? Whereas on the right, how to pass is blended in. Small faces, no emotions.
So big faces on the left, small faces on the right, not ideal. Unpolished graphics, right? On the left, showcases the top five paddles.
On the right, the guns look very similar. Small text on numbers. You see on the right, too much text, too much clutter, doesn't follow the rule of three.
Whereas on the right, big text, big numbers. So it's easy to see. On the left, you know, yellow arrow.
Whereas on the right, you know, red arrow doesn't really know where it's pointing towards, right? Unesthetic imagery. So on the left, right, that's a video on Italy tour.
So you can see Real Italy and you can see all of Italy. Whereas on the right, you know, kind of like the text is on his face, right? Background doesn't fit the old money aesthetic.
And it's really taken away from the thumbnail.
Okay, so those are a couple of thumbnail mistakes. Let's talk about A -B testing thumbnails. So why A -B test thumbnails, right?
Because this is a good opportunity to revive flatline videos, especially if you know that a video is good, how can you revive it? So, and A -B testing thumbnails can also boost videos that are already performing well. So if a video that's performing well, you want to perform multiple tests on it.
And we're going to talk about how you can go through these multiple tests.
The thing with A -B testing is that if you screw this up, you can kill a video's growth, right? You can take a video that's performing well and destroy its momentum. So we need to do this right away.
So that's exactly what I'm going to show you.
So what is A -B testing exactly?
One of the things that you need to understand before you test is that people will, you know, kind of come up to me and be like, hey, what's the baseline in terms of click -through rate, right? But really the only baseline for click -through rate is your channel's click -through rate, right? So if you haven't published any videos yet, right, I take it as 5%.
That's the baseline that I want to have. But it's important to know your channel's average click -through rate. So it's the percentage of people who click on your video when they see it, right?
And knowing that your click -through rate changes depending on how old the video is. right and find out what your average click -through rate is for the last 90 days and specifically in this case in the last 90 days and that way if one of your videos exceed your average click -through rate in the last 90 days then you can be like okay i'm not going to touch it i'm not really going to ab test it but if it's below then okay it makes sense to swap the thumbnail it makes sense to do an ab test in that case cool let's talk about how to find your average ctr step one is to analyze your last 10 videos so you can literally go into the analytics of your last 10 videos find the average click -through rate there right you can record the click -through rate at the 24 -hour mark at the 7 -day mark at a 30 -day mark and then you can see the performance of the average there so for example if your average 24 -hour click -through rate is 14 7 -day is 9 30 days uh seven percent then you already have benchmarks right at the 24 hour mark you can see okay if it's below 14 then it needs a significant bunk if it's at the seven day average then it's at nine percent then you need to if it's below nine percent then you can a b test it but if it's above then you don't really touch it and in the in 30 days if it's below seven percent then you go ahead and test it but if it's above seven percent and it keeps staying above seven percent then you don't really touch it
So how to set up an A -B test, right? Over here, whenever you're uploading your thumbnails, right? You can go to YouTube Studio, scroll to the thumbnail section, and then there's this test and compare button, right?
Which will allow you to test titles, thumbnails, all your title and thumbnails together.
So in terms of a couple of testing rules that I always follow, right? I like to test it with code traffic. So if you only test with subscribers, right?
Your results will be skewed because your subscribers are already like loyal to you. Okay. And really I want to focus on testing elements with huge impact, right?
So minor details like changing the color of a t -shirt isn't probably worth testing, right? You can test complete revamps, right? And ideally you test one variable at a time.
Testing one variable at a time means that you don't test the title and thumbnail together. That's usually a big mistake. What you want to do is you want to test the thumbnail first and then you want to test the titles, right?
So testing one variable at a time, do not swap too many variables during a given test. Okay, the truth is YouTube's built -in testing is fine, but... So there's also this tool called ThumbnailTest .com, which I recommend that you also use if you're a little more on the advanced side.
If you're like, hey, I don't really want to pay a subscription, then you should just use native YouTube Studio A -B testing. And the reason that we use Thumbnail Test is that it allows hourly testing where you can do multiple variations simultaneously. So you can test with five different variations.
and you can look at the click trade times avd over here so it's a little bit more advanced analytics to figure out what the winner is so once again right you have your 24 hour check if click rate is below right in the previous example with 14 you test if it's above you leave it alone same thing okay if it's below seven day average you test if it's chilling then you don't touch it 30 day is the same if it's below you know you can test the title now right but you know at the 30 day mark because you've already tested the thumbnails you test the title okay a couple advanced strategies in terms of new uploads right you guys really love this stuff so i want to give it all to you right so the step one is to upload with the best thumbnail okay and you can set up an hourly test four to five variations right away And then step two is to monitor the first complete cycle for click -through differences.
And step three, if there's a clear winner, you can end the test early and then implement from there.
uh rehash strategy for older videos right this is a little bit more on the advanced side where if you have older videos if you have a back catalog you want to rehash those right so when do you rehash right if you have underperforming videos or you already have successful videos or you have outdated visual elements rehash means to redo your older thumbnails okay so same thing test settings right new uploads you want to do hourly testing for the first 24 to 48 hours with thumbnail test ideally if not you can just use youtube studio natively uh any rehashes you can do daily testing for two weeks and then you can you'll do a abcd format rotation okay and kind of the most important elements are your click create times your avd which is what youtube studio looks at and what thumbnail test looks at so for a long -term strategy for your thumbnails, right?
You want to monitor all of your videos continuously. And if click rate drops below average, you want to test it and always understand that YouTube is evergreen and your old videos can be revived with a new title and thumbnail. And now that we've gone ahead and generated the thumbnails for the videos, the next step that you will need is to actually go ahead and create the videos.
I'm going to show you exactly how you can go ahead and... create your own production pipeline. I'm going to show you a pipeline that I myself have created inside of subscriber video.
And all of this is built using Claude over here. Just to give you a sample video of what it can go ahead and create. Let me just play one for you right now.
Denser sitting in my neighbor Pete's side yard, caked solid with a summer of dirt. Pete was staring down a tech bill pushing $300. And the whole problem was sitting right there on the outside of the unit, not buried inside it somewhere.
My name's Gary. and i've spent 40 some years working every trade there is mechanic to roof for the general contractor cleaning coils exactly like as more times than i can so you can kind of see this is an ai avatar that is speaking and you can already go ahead and see inside of this video it's basically you know a bunch of images b -roll avatar that is being sourced from the entire internet to put a video pipeline like this together so this is just one of the types of videos that you can go ahead and create with claude by putting together the pipeline in terms of different functionalities that we have so really the big goal of this you go ahead and add your channel type you add a title you add your script and over here you can go ahead and add any images that you want to include inside of your video if you have any you know type of media that you want to add inside of your video you can select your type of avatar you can select the quality of the motion you can select from over you know
here we have 11 labs and in world baking that is over 16 000 new different types of stock voices you can train on your own voice so that you can go ahead and replicate yourself inside of a video you can create you know fast bureau type videos you can create you know with slides with you know vivid type of content cinematic right and faceless news you know property listing type videos you can pretty much create any type of long form video inside of know this pipeline and really that's the entire goal here is you know once you actually have your script writing automated and you have your thumbnails automated this is the next step that you will want to automate which is actually the video generation even if you want it you could create a pipeline like this you can upload your recordings over here so this is for example for a client you can see that this video is 20 minutes long and then it cuts it down to 18 minutes long so you can see it cuts down the length of the video just by removing all of the ums ahs pauses right and your processes like these will actually save you a bunch of time whenever you're gonna go ahead and create the videos you can also within this platform go ahead and create an ai avatar that you want so let's say you want to generate a photo of a friendly guy
They're in their early 40s. They're a man. They are American.
Here in this case, blonde hair. Long, healthy complexion. A little rugged.
And then you can see it goes ahead and creates these for you. You can just go ahead and click generate photos. And man, like you can go ahead and pick what you want to do based on this, right?
You would just give it a little bit in order to generate an avatar. So you can even use this as an avatar that will replace you on YouTube. You don't even have to record.
videos anymore with a pipeline like this that you can go ahead and create you can also clone your voice we have a voice tab over here you can go ahead and clone your voice and best of all we actually have an api so you can go ahead and generate these videos all of it inside of your cloud so i'm going to show you how to do that plus you know some of the data validation stuff that you can go ahead and do with something like this will give it a little bit more in order to generate this video man you can see over here it is just slowly generating there's this man over here that looks like super duper realistic it actually looks like a real human and this is just with you know really advanced prompting over here this one he looks i think he looks pretty real this guy looks pretty real too So these are like super hyper -realistic images of people.
And you can kind of see like it, it really looks like it mimics a certain individual. And really, if you were to build a pipeline like this, you could go from idea script thumbnail. That's all already generated to actually the full video being generated.
And it will cost you about $20 to $30 per video, which is just crazy in terms of a long form video, because previously it would take you hours and hours and hours where you have to wait for freelancers in order to edit the videos. But, and then the freelancers have a ton of mistakes, but now you have all of this in control of yourself.
If you are able to prompt Claude to actually generate something like this for you. So let's talk about how we can do some of the video generation. ourselves inside of cloud as well as some of the different pipelines that you can go ahead and build using cloud with this okay so basically what you'll want to do whenever you are going to go ahead and put a pipeline like this together is actually scope out exactly how you want to do it so you want to scope out your exact approach of how you are thinking about executing something and then you want to nail it down in terms of you know writing it down what is the exact thought process and also nailing down what are the different tools that you can go ahead and use i'm going to give you some of the tools that we use inside of automating this inside of claw code that you can go ahead and use today right and really it's super duper important to just nail it down like granular, exactly what are different steps, right?
And it's like telling a human you want to give it specific data so that it can actually execute on these things for you. So let's say we were at the start of the script. Before you even go ahead and write like how you want to automate it, you have to be very clear what is the specific type of video that you actually want to go ahead and automate.
Let's say... it's a simple slideshow video this is the most simplest type video right i have a channel that's doing 2 000 views every 48 hours just completely automated pipeline that's uh helping me promote subscriber what you want to do is let's say the type of video right is a screen share of a slideshow and basically you want it to be going over and clicking over the slideshow let's say in a five minute video you have 50 slides and you click click click and it's essentially how motion graphics look like so When you go ahead and write the script, you're going to want to think of, okay, what are all of the different components here?
So we have the voiceover that needs to be generated. We have the slides that need to be generated. A screen recording that needs to be done.
We have the rendering of the video. So the rendering of the video is basically turning it into real life. You have to think about how you're going to render a video, where you're going to render the video, how, where, what tools that you're going to use for each of these.
And rendering the video is the very last step. So we've gone over the script. So once we have the script, basically then we want it to create the voiceover.
So what tool are you going to use? Let's say we use 11 Labs, right? 11 Labs is like a voiceover tool.
You can literally use any voiceover tool. Is it going to be your voice? Okay, yes.
If it's going to be your voice, then we need to use voiceover clone. And then now that we have the voiceover, how is the slides going to know exactly what the timestamps are? So not only do you have to write the script, you have to create a voiceover.
And then once you created a voiceover, then you have to do timestamps transcribing. Okay, once you have the timestamp transcribing, then you're going to want to go ahead and this one will be using ChatGPT Whisper, which is one of the tools in order to use timestamp transcribing. Once we have the timestamp transcribed, then how do we decide how many slides there are?
So it's literally thinking about it in like super duper sequentially. So then it will chunk it into beats. Let's say we want to target video length is six minutes.
20 25 to 30 slides but the beauty of building these things with claude is that if we're not sure how to do the screen recording we can go ahead and ask claude how we can do the screen recording okay and then once we do the screen recording what are we going to do within this so like how where and what and also is it just going to be slides okay is there going to be videos is there going to be like where are the videos going to be found What happens if it doesn't find any?
These are like all of the questions that you'll want to find so that, you know, literally there's all of these different steps that you're going to have to go through and you want it to solve for each of these. So for example, if you want to source videos, you're going to ask it like, okay, like where can we find these videos?
It could be like, okay, we want to use YouTube to find these videos, or we want to use like storyblocks in order to find these videos, for example. What happens if you don't find any? If you don't find any, then it should fall back to XYZ or it should be able to find it and you have to build the logic for that.
And like, how are you going to do the screen recording? Or like, how are you going to go ahead and build the slides? So then I'm going to want to go ahead and just search PowerPoint Claude scale.
awesome skill a slide powerpoint designer okay awesome skill okay then i can go ahead and just clone this git and then i can go ahead and use this so let me just go back to my board i'm just like slowly mapping this let's say we want to do it of this and then we want to test it to see if it's good okay how are we going to do screen recording screen recording clawed skill screen recording clawed skill awesome skills .ai this one's really good okay that sounds exactly like what i want so i'm just going to go ahead and paste it over here into something that i'm using and then let's say in the video I want the first 30 seconds to be a video.
So you can kind of see I'm just like mapping like high level how I think about how to do this. That way I can go ahead and prompt Claude on exactly how to go ahead and create this. And I'm just going to go ahead and open Claude.
I subscribe a pipeline. Let's start a new branch from main. In terms of like the refinement loop, high level, what I want to do is I want to plan, plan, and then execute.
And then check the work, run tests and review the tests. And then I want to go ahead and plan again. Like this is literally the loop that you would run it through.
And if you go through enough of these loops and you give it enough of these feedback, it will be a lot easier in order to go ahead and create videos. Even like reviewing the test, you can use Gemini Flash to watch the videos for you so that you don't even have to go ahead and watch the videos. Of course, it will be good to have human input on exactly how to watch the videos.
But, you know, this is this will be good and this will be important. And once again, you can do this for any type of video. This is just a slide type video that I'm going to demonstrate.
But you can kind of see. video that i showed you it's like using you know story blogs using image sourcing in order to put the entire video together you can definitely do that too first i have to go ahead and create the plan in terms of a rule of thumb We only use Fable 5, which is Claude's best, strongest model in order to plan exactly what you're going to do inside of your code.
And then we use Sonnet 5 on low effort in order to execute. As much as possible, we want to use the Chinese models in order to reduce the cost as much as possible. Chinese models is basically companies from China that are attempting to reduce the cost and get people to use them instead of the frontier models.
Because if you have used Claude before, if you've used Sonnet before, you will know. that's actually pretty expensive in order to use these models. And because it's so expensive in order to use these models, it's a lot of times just better in order to go ahead and create videos in terms of my favorite ones.
You know, it's Kimi K3, DeepSeek V4 Pro, and even like DeepSeek V4 Flash. Going ahead and playing around and testing with these models to see what can actually give me like the best output possible. These are some of my thought process that I use to go ahead and build whatever pipelines that you want.
If you go through this process, right, you can literally build any video from scratch, right? You can automate full faceless videos. You can automate talking head videos with all of these different video models, image models, to go ahead and create a pipeline that you...
think is like really, really worthy. I have this entire pipeline. I'm going to go ahead and just talk to Claude.
He started a new pipeline called Video Test. I'm going to want to give it a role and I'm going to tell it exactly what to execute with. So you can see over here, I'm just going to switch to Fable 5.
High effort. I'm going to want to tell it to run like an autonomous testing loop too. I'm going to give it a role and give it a goal.
In terms of what I'm trying to do, I am looking to automate exactly. this specific video pipeline. I want to call this pipeline a video test.
So how this works is the input that I'm going to give you is a script. And what I want you to do is I want you to create the voiceover using 11 Labs API key, which I'm going to go ahead and give you. What I want you to do is I want you to do timestamp level transcription using ChaiGBD Whisper in order to transcribe it so that I can get like second by second level chunks.
And then with this script, I want you to chunk it into beats. I'm going to give you a plot scale on how to create slides. And I'm going to give you a Claude scale.
Green are called those slides. That's pretty much what I want this demo to be. I have...
chunked it into beats like basically i want you to chunk the script into you know 25 to 30 slides worth of beats and the slides should represent and support what is going on inside of the slide and then i want you to also add images onto the slide so that it looks visually interesting i'm also going to give you my specific style guide on how i want the slides to look i want you to go ahead and create the slides create a skill for creating the slides and once you've gone ahead and created the slides i want you to use a screen recording skill to go ahead and screen record exactly how the videos look.
When you're doing the screen recording, basically I want you to cycle from one slide to another based on the timestamp. transcription and matching it to the beats and the slides that is being shown in terms of the text on the slides i want it to be you know pretty minimal and have you know some images so that it looks good and i want to use the screen recording scale and then i just want you to render the video completely locally eventually you know we will take the rendering of the video and then put it onto the cloud In terms of the hook inside of the first 30 seconds too, I want you to go ahead and source videos that I can use inside of the first 30 seconds.
So I want you to look at the second by second transcription, look at the first 20 to 30 seconds, and then see what is the best fit video that you can go ahead and use in order to make the video a little more interactive and really, really look good. In terms of the models that I want you to use. I want you to plan out exactly how we will go ahead and execute this using Cloud Fable 5.
I want you to execute on the plan using Sonnet 5 on low effort. And then I want you to check the work that you have done. Like you're a computer scientist, you can go ahead and review the work.
And I want you to go ahead and run some tests, run a smoke test with a couple of titles of videos. And I'm going to give you some scripts. I want you to use Gemini Flash in order to watch these videos so that you can review the videos.
And then once you have reviewed the videos, figure out what's wrong with the videos and then go ahead and plan. And I want you to go through this loop multiple, multiple times until you can get me a 10 out of 10 video that you have watched using Gemini Flash. Now, what I want you to do.
And this will be really, really useful. Create this plan and figure out any unknown unknowns that might potentially stop you from getting a really, really strong output in terms of the video that makes the video actually look really, really good. I want you to go ahead and do that for me right now and ask me any questions about how this pipeline will operate and maybe what models that we will be using at each individual step here.
And you can see it just populates this entire thing. So then I'm just going to click enter here. When I'm speaking it out loud, I'm using this tool.
It's called fluid voice, right? I have no affiliation of them. This is really, really nice because number one, it's free.
You can just download it. It's open source. It's not like whisper flow.
And I can just kind of talk out like my thought process. And you can kind of see when there are imperfections, when I'm saying out loud, my thought process, it actually allows it to kind of understand what I'm thinking. So it's much better to speak it out loud rather than typing it.
It's also a lot faster. And one of the things that I typically do when I'm creating pipelines like this. is that also I ask it, what are any unknowns and knowns that might potentially stop you?
So if you guys know the concept of this, known knowns are things that we are aware of and understand. These are like the easiest things in order to automate because we know of it and we understand it and we can just tell it, automate it for us. Known unknowns is that we know that we don't understand certain things.
So then we can ask it, I know I don't understand this. Help me answer it. That's what people use HGBT for.
That's what people use Google for in order to understand certain things. And then there are unknown knowns, right? It's things we understand, but we are not aware of.
this is also important to uncover and unknowns unknowns are things we are neither aware of nor understand this is especially true when you are going ahead and asking claude or chb t to go ahead and program something that you don't already know of because you don't know of it and you don't understand it you can't actually prompt claude in order to go ahead and do that so what you want to prompt it to do and this is especially really good with cloud fable 5 is you want to tell it to identify what are the unknown unknowns identifying what the unknown unknowns are will actually help you figure out when you are going to endeavor and to something that you don't really understand, it will poke holes and really find an intelligent way of figuring out.
what are really the blockers that will stop you from going ahead to actually create these videos so these are called unknown unknowns we are going to ask claude to help me identify what are the unknown unknowns that are going to stop me from getting the desired output that i really want it's important to tell it hey figure out what are my unknown unknowns and figure out what are the blockers that will stop me from creating a video that is really really top class and that will make this video pipeline fail this could be things with cloud rendering this could be things with voiceovers this could be of transcriptions by the way this is also useful for application in literally any field to make you really aware of what are the gaps there i'm just going to let this pipeline cook and we'll see what questions it asks us there we go okay what is the timing source for the slide sync okay we're just going to use 11 labs how do we capture the slides real -time screen recording drift frames okay so over here i just totally forgot
to go ahead and feed it what the skills are. So I'm just going to feed it to two skills that I just showed you. Use these skills.
One is for slide generation. The other is for video generation. And the other is for screen recording.
Clone those kits. Okay, where does slide images and hook comes from? Let's just tell it to use stock APIs.
Okay, where does the video test pipeline code live? Okay, so it just lives here. We're going to come back.
when it has something and right now what i'm going to do is i'm going to show you a sample video that is actually generated in terms of like this is the architecture of the pipeline that i want let's just watch this video to see what it comes up with anti -gravity is free right now zero dollars while cursor costs 20 a month and that gap is real money if you're running a team of 10.
anti -gravity also holds the only published swe bench score in this fight 76 .2 percent verified All three tools just hit major releases in 2026, so I'm running Cursor vs. Antigravity vs.
Windsurf through five head -to -head tests to see who actually wins. The free price tag sounds like an obvious win, but whether that $20 a month saving actually matters depends on something we don't settle until the very end. My name is Brian, and I'm in these tools 12 hours a day, so you don't have to be.
Just right now. while Cursor's cheapest paid plan is $20 a month, and Windsurf matches that exactly. All three top out at $200 a month for their premium tiers, so the high end is a wash.
The one caveat worth flagging, I couldn't find Antigravity's team pricing anywhere, and on per seat cost, community reports put Cursor at $40 per user, versus Windsurf's $80 base, plus $40 per seat. Antigravity takes test one. And it's not subtle.
The pricing looked like it'd be pretty even across all three going in. And then anti -gravity is full. And you can kind of see over here, right?
It just goes through a bunch of these different slides. And you can see like the main use case for this that I had it for is building these software comparison type videos, right? And of course you can go ahead and use something like this to go ahead and generate software comparison videos, even like educational videos, because something like this slides is used.
to teach literally tons of concepts the most viewed video on this specific channel a bunch of google slides being putting a bunch of gamma slides being put together and just me reading over those gamma slides so as long as the information is good as long as it's interesting you can kind of see over here that the content actually comes out to be pretty good something that i'm very very happy with you can go ahead and create slides like this and i think this is really really cool to kind of see that even now something like screen recording and slides you can go in and automate courses like b -roll You know, a lot of you would have seen being automated by a lot of different software companies, my company included, but you can go ahead and create pipelines like these yourselves.
I think it's, you know, pretty effortless and it takes a little bit of time in order to get right, but actually is pretty, pretty. good okay now before we end off i want to go over some pretty interesting niche use cases that really i am using to automate a lot of the processes with my youtube channels some recent new cloud features that i just had to add inside the tutorial that i think is just really cool to see what we can even automate nowadays with cloud co -work i don't typically use co -work but i use it for this specific function where you can go ahead and if you click the plus button there's this thing called record a skill and basically what this is if you hold a scale of you going over exactly how you are doing something and you talking over it, go ahead and see what is happening on screen.
It'll document all of the exact steps and then turn it into a skill that can help you do the repetitive tasks. So one of the repetitive tasks that I simply hate doing YouTube is uploading videos because it's like, Hey, I run so many different channels. Why is it that I have to go ahead, click on these buttons, upload these videos myself.
It doesn't make sense. This is something that AI should do. But one of the problems with this is that when you're using the YouTube API, which is what a lot of people will do.
Number one, most people don't have access. It's capped at six videos a day that you can do. You know, there are different groups of people who are thinking about it, but some people will say that the YouTube API will actually suppress how well your videos perform on browse.
YouTube thinks that it's automated because it's done with the API and there's people who are abusing it. So that's why using the screen record scale, it's really useful in order to just upload a ton of these. Okay, so this is how it works.
You just click start recording here. and then you just want to talk over it so you can kind of see it's capturing all the different steps. And once you're done, you can go ahead and click done.
So I'm just going to do this on a throwaway YouTube channel. The first thing that I will want you to do is I'm going to just go ahead and upload whatever video that I have within here. You can kind of see that when you're over here, you can just go into content, right?
And you can filter. by the videos that are inside of the draft once you filter by the videos that are inside of the draft kind of see that this video will appear inside of the draft what i want you to do is go onto this channel the name is natalie pierce is to click on this specific video filter the videos by the draft click into the specific video and what i want you to do is take this title right and go on to chat bt and then i want you to tell it hey fix this title and give me a description for it and then what i want you to do is just copy the title i'm going to just copy it and then you can paste it over here and then inside the description you can copy all of this and you can paste it over here and then for the thumbnails i want you to click upload file and what i want you to do is go into downloads and then find the name of the thumbnail that is correspondent with this specific video.
So let's say it's this specific video. The name is correspondent, right? This one is not exactly correspondent, but make sure that it's correspondent.
I want you to scroll down. I want you to click on it's not made for kids. And I want you to scroll down, allow automatic places and allow automatic concepts, right?
I want you to cancel all of this. We can skip the tags because tags are not important. I want you to click next.
I want you to click import from video. And I want you to click this. If there is an error here that you can kind of see.
We want to make it shorter over here. Click save, click initial check and then visibility. And then I want you to set it.
private right we don't actually have to go ahead and publish it because i don't want you to publish it i want to go ahead and double check before publish it and then click save over here and voila you would have done the uploading of this specific video i want to be able to replicate this for all of the other channels that i'm doing so once you click done over here you can kind of see that it's going to start processing this i already have a magic scale let's just stop it i don't want it to update the scale because this scale is something that i have been using already right but you can see over here If you look at the skills that I have, it's published YouTube draft.
I already have this skill that's being used and I'm just using it to upload all of my videos. You can kind of see during this chat where basically it will just go on in here and then it will go ahead and upload those videos for me. So I think that is one of the really cool use cases that you can have for this cloud co -work skill, especially if uploading videos is one of those things that, hey, you're thinking I'm going to hire a channel manager or a VA in order to go ahead and do that.
Right now, you could just go ahead and automate that. The second thing that I want to show you how to do is that you can actually go ahead and create. really, really cool dashboards.
Let me just pull it up for you. you can see over here basically dashboard that i have for channel intelligence basically what it's tracking is what are my competitors what are the main competitors i have 93 channels tracked what are their lifetime views what are their median views of video what is their outliers there is a bunch of these different filters that i have and more importantly it creates a bunch of playbooks so it looks at what the competitors are doing and then it's finding what is similar between what the competitors are doing and it creates playbooks over here so it's like hey like why do these videos work and you kind of see it labels those specific videos to why those work and then it ranks those as to okay what are the best what are the second best within this you can kind of see this and also within this if you scroll down a little bit more you can see that this gave me an audit of the channel so it tells me oh like ideation and topic selection needs help title and packaging this is what i would change thumbnails editing retention or this is what i would change channel positioning this is what i would change ranks these playbooks and then it actually lists all of my competitors within here it actually finds all of my competitors for me and i have them all inside of this dashboard right you can see
I can find their niche, the number of subscribers they have, how old is their channel? What is their views per video? What is their outlier?
And what is the best sample video that they have? So what is their best performing videos? I have an overview of my entire niche just from a dashboard that I have here created inside of Claude.
And then it also creates a thumbnail wall for me to see what are the competitors doing and all of the different samples within there. I can go ahead and filter these. And also this is just a sample dashboard, which I think is really useful when I am trying to look at certain pieces of data.
Maybe I cannot. see and that i'm missing that if i really had access to this piece of data it would change the way that maybe i do a certain video maybe i don't do a certain video just because i'm able to visualize something like that so i'm going to give you guys the steps in order to do this in terms of how we get clawed to interactive youtube what i do is i use the subscriber mcp inside of subscriber we have an mcp that will allow us to immediately go ahead and talk directly to youtube if you go into your channels and your cloud settings.
If you go onto extensions or connectors over here, and you can kind of see all of the different MCPs that I have connected over here. But over here, we have the subscriber connected directly. If you want to add an MCP, you can add a custom connector over here.
And then all you have to do, open your Google Chrome, you go to subscriber .ai slash MCP, and then you can copy this MCP over here and paste it over here and give it a name. It's your subscriber. It'll verify your subscriber login, and then it'll be able to connect your MCP directly to it.
And you can kind of see all the different functions that it has because the subscriber has 40 million videos and pieces of data. You can get research video. You can get script project within this.
You can see, get my project, list research bookmarks, comment a voice profile. So you can go ahead and change the voice profile within this. create project you can also generate ideas you can mark project notifications red promote an idea save a bookmark save a template literally this mcp is literally all that i needed in order to build a dashboard like this i was able to create a prompt that will allow me to create a dashboard like this i'll go ahead and link this prompt in the description over here so you can go ahead and use this you can see this prompt is just going on forever and ever and ever but it allows you to really create a dashboard like this that i think really allows you to really visualize a lot of these pieces of data, like what thumbnails, what specific playbooks.
And because Claude is just really good, or AI in general is really good at categorizing these pieces of information, you can kind of visualize it. And I think this is just really cool. You can keep prompting it so that you can go ahead and refine this.
It's just a simple dashboard that I got from a single prompt that allows me to see who all of my competitors are. And now one of the last things that I do with Claude is that I go ahead and tell it, hey, how can it actually improve my videos? Because Claude has...
access to youtube now a true subscriber you can actually have your retention data and your average view duration data so it's basically has access to data as to how long people stay and the specific script of your video because it has access to that you can kind of see like hey like what are the video ideas that you've done what are the length of the videos where is there like problems in terms of the scripts and it's able to audit all of these things and you see i just told it to create a report like this for me it breaks down the script structure that i have where it's like preamble test one test two test three test four test five recap and then it pulls all of this.
And then line by line, the first 75 seconds, the leaks off there. GemNotch 3 .6 flash, 6 .7 times cheaper per token than Fable 5. It's given me all this in -depth analysis on the scripts.
And then it's telling me like, hey, my AVD for this specific channel is one minute and 30 seconds. I'm trying to get it to two minute and 30 seconds, right? What can I do in order to do that?
It's giving me the math, giving me like, oh, what other different templates that we can go ahead and use. So it's giving me all of that, like super duper in -depth. You can just talk to Claude and you can tell it like, hey, you have my specific retention data.
What is the specific... changes that you would make to the videos in order to improve it go a second level deeper and discover any unknown unknowns that i have within like the script writing process that is potentially leading to the abd just being so so so terrible and you can just prompt it like that right and you can literally just talk to it back and forth until it is able to give you an answer that you're satisfied with the entire process like hinges on this it's your plan execute checking the work run test and then review the test it's going through this flow as many times as possible until you can get a pipeline that you are really really happy That is 90 % of what to do with Claude combined with the YouTube information that we have covered inside of this video.
Once you combine like subject matter expertise with knowing how to automate of AI and combine those two, that is what really makes creating content like this all on video pipeline or thumbnail pipeline are much more effortless than I think. a lot more accurate. And yeah, this is the end of the video.
Thank you guys for watching. If you guys have watched the entire few hours, you are officially a master at Claude and YouTube. Hopefully that you learned a thing or two.
And more importantly, that you actually apply this knowledge to go ahead and grow your channel. So now a couple of ways that you guys can go ahead and implement this a little bit further. Firstly, if you guys don't already have access to subscriber .ai, you can click the link in the description and you can connect your AI agent to it for free today.
Just click the link in the description below. We have an API, which will allow you to create scripts, create thumbnails inside of Claude. We have an MCP, which will allow you to access all of these pieces of data and also build dashboards like the one that I just showed you within here.
And we also have a video pipeline that will allow you to automate your video creation from, you just give it a script, click a couple of settings, and then your entire script will be spit out that way. And leave a comment below on exactly what you're going to be building with Claude and YouTube, because I want to hear exactly what type.
type of videos that you want on all of the different types of things that you can build with Claude, because I'm doing this 24 hours every single day. I'm just creating this course so that I can bring these learnings over to you. And also, if you want to watch more of my stuff, click over here so that you can learn exactly how to grow your YouTube channel with banger YouTube scripts and AI.
And if not, thanks for watching. I'll see you in the.
The Hook
The bait, then the rug-pull.
The promise is stated flatly in the first sentence and the filter arrives forty seconds later: if you cannot sit still for five minutes, you are not the audience. What follows is not a tips video. It is seven modules, each one demonstrated on screen, that together describe a channel as an operating system rather than a series of uploads.
Named ideas worth stealing.
04:16concept
Audience-first niche filter
- 55+ demographic
- Tier-one countries
- Target RPM near $8
- Simple editing tolerance
- High purchasing power
Choose who you serve before what you cover, then check the niche against age, geography and expected RPM. Topics are treated as an output of the audience, not an input.
Steal forany content business where the same effort pays very different rates depending on who consumes it
07:41list
Three ways to find a niche
- Mine Instagram and TikTok for formats not yet on long-form
- Port a working format to another tier-one country
- Add a credible face to a faceless niche that requires trust
Three arbitrage moves, each based on the same logic: find demand already proven somewhere, then move it to a market where nobody has served it yet.
Steal forproduct positioning and offer research, not just video niches
17:05list
Channel warm-up protocol
- Watch videos on the account before posting
- Ship one real test video
- Then ten videos, one every two days
- Read the outliers at day 20
- Kill what failed, double down on what broke out
A fixed 20-day testing window that treats the first ten uploads as a data-gathering run rather than a launch, with the decision point deliberately placed at the end.
Steal forany launch where you would otherwise judge success off a single piece of content
23:55list
Five AI avatar realism rules
- Realistic background with depth
- Human, believable proportions on the rule of thirds
- Life and reflection in the eyes
- Imperfect skin texture, not smooth
- No hands showing
A checklist of the specific tells that make a generated presenter read as fake, ordered by how often each one gives the image away.
Steal forauditing any generated image of a person before it ships
29:54model
The nine-stage video board
- Idea
- Packaging
- Script
- Voiceover
- Visuals
- Editing
- Review
- Scheduled
- Published
Every video occupies exactly one stage at a time, which makes bottlenecks visible as column pile-ups and makes ownership unambiguous.
Steal forany repeatable production process where work stalls invisibly between steps
39:40concept
Plan 80, execute 20
Most of the operator's time goes into interrogating the plan, with the model instructed to ask questions until it is 95 percent certain and to critique its own plan from several named expert perspectives. Execution then runs unattended.
Steal forany agentic build where the temptation is to one-shot the prompt and fix it later
46:59model
Topic + format = viral idea
- Viral topic: what you talk about
- Viral format: how you present it
- Both required, neither sufficient
The worked example pairs a topic with proven search demand against a format proven in a different niche, on the logic that a great topic in a bad wrapper never gets clicked and a great wrapper around a dead topic has nothing to sell.
Steal forheadline and offer construction generally
48:41list
Six places to find high-click ideas
- Your own outliers
- Top channels in your niche
- Adjacent niches, same audience
- Bigger niches you can borrow down from
- Unrelated niches where only the title ports
- Whatever is top of mind this week
Ranked roughly by confidence. Your own proven videos are the safest source; a format lifted from an unrelated vertical is the riskiest and usually only the title survives the move.
Steal fora standing research routine rather than a one-off brainstorm
1:07:29list
The eight core video templates
- Challenge
- Documentary
- Listicle
- Storytelling
- Educational
- News
- Commentary
- Review
Almost no video is a pure template. The useful move is naming the blend as a rough percentage so the writing instruction has a shape to follow.
Steal forbriefing any writer, human or model, on structure before style
1:13:28list
The seven-step script chain
- Deep research via API
- Verify the research
- Compile a research report
- Outline against the template
- Write the draft
- Cut repetition
- Humanize in 500-word chunks, then voice check
Each step is a separate call with one job. The stated design goal is to reproduce what a human researcher and writer would actually do rather than compressing all of it into one prompt.
Steal forany long-form writing workflow currently running on a single prompt
1:36:31list
Five curiosity-gap thumbnail categories
- Moment
- Story
- Result
- Transformation
- Novelty
Each category opens the gap differently: dropping the viewer into a scene, implying an unfinished narrative, showing the outcome they want, showing before and after together, or showing something they have never seen.
Steal forad creative and email subject lines, not only thumbnails
1:42:30acronym
The three C's of thumbnail design
- Content: which elements appear
- Composition: symmetrical or thirds-based placement
- Contrast: brightness, saturation and hue against the feed
A design order of operations. Decide what is in the frame, then where it sits, then how it separates from everything around it.
Steal forany visual asset that competes in a crowded feed
1:55:00list
The three thumbnail tests
- Clarity: are all elements legible at the smallest size
- Contrast: does it separate from competitor thumbnails
- Glance: can someone explain it after two seconds
A pass-fail check applied before upload, with the additional rule that if you cannot state the curiosity gap in one sentence, it is not strong enough.
Steal fora pre-publish checklist on any creative asset
2:03:00model
CTR baseline and test cadence
- Record CTR at 24h, 7d and 30d across your last 10 videos
- Below the 24h average means repackage now
- Below the 7d average means A/B the thumbnail
- Below the 30d average means test the title
- Above average means leave it alone
Turns testing into a decision rule tied to your own history, which also prevents the common failure of disturbing a video that is already outperforming.
Steal forany optimisation loop that currently runs on gut feel
2:23:20model
The known and unknown matrix
- Known knowns: automate directly
- Known unknowns: ask and learn
- Unknown knowns: surface what you already have
- Unknown unknowns: make the model enumerate them
Applied as a prompt pattern before building anything outside your own expertise, on the logic that you cannot specify a requirement for a failure mode you cannot name.
Steal forscoping any technical project in a domain you do not know
2:20:00model
Plan, execute, check, test, review
- Strong model plans
- Cheap model executes on low effort
- A reviewer pass checks the work
- Smoke tests run against real inputs
- A vision model watches the rendered output
- Loop until the output scores 10 out of 10
A closed loop with an automated grader at the end, so the operator is not the bottleneck on every iteration.
Steal forany generative pipeline where quality is currently judged by hand
How they asked for the click.
VERBAL ASK
2:38:01product
“If you guys don't already have access to Subscribr, you can click the link in the description and connect your AI agent to it for free today. And leave a comment below on exactly what you're going to be building with Claude and YouTube.”
Soft-sold throughout rather than saved for the end. The tool carries most of the demos, and the ownership is disclosed once in passing around the script-writing module, well after it has already appeared in four earlier sections.
MENTIONED ON CAMERA
Visual structure at a glance.
