Modern Creator
Sam Gaudet · YouTube

How to Make Insanely Good Content (with AI)

Five principles for using AI on content without producing slop: feed it your own data, split the work 10/80/10, train it instead of prompting it, hold the standard, and let an agent work ahead of you.

Posted
2 days ago
Duration
Format
Talking Head
educational
Views
3.6K
83 likes
Big Idea

The argument in one line.

AI content only stops being slop when you feed it your own experience plus what is already working online, train it on your rules once, and let an agent draft ahead of you.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • A creator who uses ChatGPT or Claude for scripts and keeps getting generic output that sounds like every other AI-written video.
  • Someone making short-form or YouTube content on a regular schedule who restarts from a blank prompt every single day.
  • A solo operator with call transcripts, notes and past content sitting around that has never been fed into an AI system.
SKIP IF…
  • You want a ready-made prompt to copy. The video argues against long prompts and the actual system prompt is gated behind a DM.
  • You do not publish content regularly, so there is no past work to train a system on.
TL;DR

The full version, fast.

Most AI content fails because people ask for ideas and read the result word for word. The fix is five principles. Feed the model your internal data (transcripts, notes, past posts) and external data (what performs in your niche) and look for the overlap. Split the work 10% ideation, 80% frontier-model execution, 10% distilled feedback. Train the system on voice rules, banned words, preferred structure and self-review instead of writing 1600-word prompts. Push back on the first draft with specific requests. Finally, run a proactive agent that drafts scripts and asks you for input daily.

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 →
Chapters

Where the time goes.

00:00 – 00:22

01 · Cold open

Names the common failure: asking AI for a viral video and reading it verbatim, then promises five principles.

00:22 – 01:48

02 · Stop asking AI for ideas

Feed the AI internal and external data and find the overlap, which becomes both the content and your moat.

01:48 – 02:48

03 · The 10/80/10 rule

Ten percent ideation, eighty percent frontier-model execution, ten percent distilled feedback.

02:48 – 04:45

04 · Train, don't prompt

Four training categories: voice rules, banned words, preferred structure, self-review. Build them from your best content.

04:45 – 06:42

05 · Hold the standard

Four feedback requests: novel and non-obvious, nuanced and observable, contrarian takes, define and redefine terms.

06:42 – 08:10

06 · Proactive vs reactive process

Install an agent that runs the first four principles daily, pulls data, drafts in your voice and asks you for input.

08:10 – 08:50

07 · Feedback loop and CTA

Retrain the system on your beliefs, not just single outputs. Offer of the system prompt via DM.

Atomic Insights

Lines worth screenshotting.

  • Asking AI for ideas gives you the same ideas as everyone else using AI, which is where slop comes from.
  • The best moat in AI content is your own lived experience, because nobody else's model has it.
  • Feed AI two kinds of data: internal (transcripts, DMs, notes, past content) and external (what is performing in your niche), then look for the overlap.
  • Content you are interested in that also matches proven demand makes creation feel effortless.
  • Use a 10/80/10 split: 10% ideation with the data, 80% execution by a frontier model, 10% clear distilled feedback.
  • Most people cannot explain why a script is bad, and until you can say it, the AI cannot fix it.
  • Stop writing 1600-word prompts. Train the system on voice rules, banned words, structure and self-review instead.
  • Set a reading level cap. Seventh grade English or lower helps people understand the concept better.
  • Ban AI tell phrases like 'it's not about this, it's about that' and em dashes outright.
  • Tell the AI to review its own work against your rules, because it defaults to the fastest path and never checks.
  • Transcribe your best content and ask the AI to build the rules that would make that output repeatable.
  • One-shotting good content is extremely rare, so plan to give specific feedback on the first output.
  • Ask for novel, non-obvious concepts when a draft shows patterns you have seen before.
  • Ask for nuanced, observable things, because AI defaults to generic theory.
  • Ask for contrarian takes, since the best performing social content often opens with one.
  • Make the AI define and redefine terms, which prompts a thought in the viewer and pulls them in.
  • A reactive process restarts from zero every day. A proactive process has the agent bringing you drafts and questions.
Takeaway

Train your AI once, feed it your life

WHAT TO LEARN

AI content gets good when you give it your own experience and proven demand, teach it your rules once, and push back with specific feedback.

02Stop asking AI for ideas
  • Giving AI your transcripts, notes and past content, alongside what performs in your niche, produces ideas no one else's model can generate.
  • Content that overlaps your real experience and proven audience demand is easier to make and harder for competitors to copy.
03The 10/80/10 rule
  • Spend your effort at the two ends: set up the data at the start and give clear distilled feedback at the finish.
  • Being able to explain why a draft is weak, like you would to a person, is the skill that makes the AI improve.
04Train, don't prompt
  • Voice rules, reading level, banned phrases, a preferred structure and a self-review step replace a 1600-word prompt.
  • Building those rules from your best past content gives you a repeatable standard instead of a guess.
05Hold the standard
  • Expect the first draft to need work, and ask for novel, observable, contrarian ideas with redefined terms to lift it.
06Proactive vs reactive process
  • An agent that pulls your data, drafts in your voice and asks you questions keeps you from ever facing a blank page.
Glossary

Terms worth knowing.

AI slop
Generic, formulaic AI output that sounds like everyone else's because the model was given no unique data or voice rules.
10/80/10 rule
A work split where 10% is ideation and data-feeding, 80% is execution by a frontier model, and the last 10% is distilled human feedback.
Internal data
Material from your own world, such as call transcripts, notes, Slack, DMs and past content, given to an AI to ground its output.
External data
Content that is performing well in your niche on social media, used to show the AI what the audience responds to.
Proactive process
A setup where an AI agent runs on its own, drafting scripts and asking you for input, instead of you starting a blank chat each day.
Hook, explain, illustrate, teach
A short-form content structure the creator gives his AI so every script follows the same four-part shape.
Resources

Things they pointed at.

01:10tooltopicfinder.ai
07:10toolClaude Code
07:10toolCodex
07:50toolGranola
Quotables

Lines you could clip.

00:00
“Most people go into ChatGPT and say, make me a viral video and then proceed to read that video word for word using a teleprompter and sound like a robot.”
Instant pattern interrupt naming a common mistake→ TikTok hook↗ Tweet quote
00:22
“Stop asking AI for ideas. It's going to give you the same ideas as everybody else that's using AI.”
Contrarian and self-contained→ IG reel cold open↗ Tweet quote
01:40
“You're creating content from your life experience, not just generic AI bullshit.”
Clear moat claim, no setup needed→ newsletter pull-quote↗ Tweet quote
The Script

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.

It's actually pretty easy to make insanely good content using AI. The problem is that most people go into ChatGPT and say, make me a viral video and then proceed to read that video word for word using a teleprompter and sound like a fucking robot. And so in this video, I'm going to break down the exact principles that we use within Dan Martell's personal brand to make insanely good content using AI without sacrificing quality.
So the first principle you need to understand is to stop asking AI for ideas. The problem with asking AI for ideas is it's going to give you the same ideas as everybody else that's using AI. And this is the reason why a lot of people end up with AI slop is because they're not actually giving AI the right data and the right information so that it gives them unique insights.
And so there's two types of data that I like to feed my AI systems. The first one is internal data. This is the data that lives within your systems, right?
Your call transcripts, your Notion, your Slack. your DMs, your own content, right? All the stuff that lives within your world.
And you want to take that data and give it to the AI system. The next one is your external data. And this is where you take what's performing in your niche on social media, other people's content.
And you can use tools like topicfinder .ai to actually find these videos that have performed really well in your niche and then give them to your AI systems. The key here is you want to teach your AI systems to find the overlap between your internal and your external ideas, because that's where the magic happens. When you can just talk about stuff that you're interested in, that you also know going to perform on social media then content creation becomes insanely effortless this is where your experience actually becomes your advantage because you can't talk about stuff that you haven't actually done and so yes go focus on doing more cool but then find the overlap between that and the stuff that's actually performed on social media so stop asking your ai for ideas and just give it what you've done and what's performing online and then make content that has overlap with both this is probably the best moat as well in a world where everybody can create content using ai you're actually just creating content
from your life experience, not just generic AI bullshit. And so you're able to actually differentiate yourself from everybody else because you're actually talking about the stuff that you've done, not just the stuff that the AI systems give you. Which brings us to principle number two of making content with AI, which is use the 108010 rule.
10 % upfront should be ideation with the system, right? This is where you give it the data, you give it all the stuff that we talked about in principle number one. Then 80 % of the execution should be done by AI and using frontier models.
If you're in late 2026, you should be using Fable, you should be using... Using Astro, the frontline models that are actually running these systems. And then the last 10 % is where the magic happens.
This is where you give it clear distilled feedback. This is where you can teach it principles to actually steer it in the right direction. The problem with most people using AI is they can actually distill what they're thinking in their brain.
When they think that's not a good YouTube script, or that's not a good short form piece of content. And what you want to do is take that information inside your brain and distill it and actually explain it to AI. Like you would explain to a person because these systems are a lot smarter than you think.
And if you can actually. explain why you don't like an idea then you can actually make it better which brings us to principle number three of making content with ai which is fuck prompting start training your ai systems actually write like you look you shouldn't be spending your time writing a 1600 word prompt what you should be spending your time on is actually training your ai systems so that you don't actually have to spend a lot of time prompting and these are the four categories that i like to train my ai systems on when i'm making content the four categories are voice rules right make sure that it actually has the rules of how you speak as an example for me i like to use contractions i won't say what is or that is i'll say what's and that's another example is i like to use short sentences whenever i'm scripting using ai and so make sure that you're teaching it that another thing i hate is anything over seventh grade level english because it doesn't actually help people understand the concept better if you are someone that's teaching people stuff on social media then you should be able to explain your concept in seventh grade or lower english the next level is banned words and phrases i like to completely ban words
words or phrases straight out, right? An example of that for AIs, it's not about this, it's about that. I fucking hate seeing that in my scripts.
And so I tell AI, hey, don't use this at all. Another example is em dashes, especially if I'm writing a LinkedIn post or a tweet, I'm going to tell it, hey, absolutely don't use any em dashes because that's a clear give of AI writing. Next, I give it my preferred structure, right?
I like to use the hook, explain, illustrate, teach framework whenever I'm making short form content. And so I say, hey, use this framework to actually make content. And the next step is the self -review.
I love to tell it, once you're done, review your own. work and notice if you've broken any of the principles that i gave you because a lot of times ai will just default to the fastest way to do something and never actually check its work and so this way you actually almost guarantee that it does the thing that you tell it to do so once you understand this what you need to do is go and find the 10 out of 10 content that you've created transcribe those pieces of content and then give it to your ai systems and say build the rules around what would make this output repeatable and that way you'll get the band words you'll get the voice rules you'll get everything that you need from your AI systems.
And that way you can train it and not prompt it with 1600 word prompts, which brings us to principle number four of making content using AI, which is hold the standard, make sure that you're holding the standard for what great content looks like a lot of times, whenever I get my first output from AI, I'm going to give it clear feedback on what I need to see more of chances that you one shot content is extremely rare.
And so what you want to do is give it specific advice on how to improve its output. And there's four things that I love to use whenever I'm telling AI to change the output. The first one is say, give me novel, non -obvious concepts.
If I look at the AI writing and I see patterns of stuff that I've seen before, I just say, give me novel and non -obvious stuff. That way it just goes and looks into the database and sees new things, stuff that I've probably never heard before. And is also non -obvious stuff that I might not have thought about already.
And it goes and finds those things. The next one is give me nuanced and observable things, not theory. A lot of times these AI systems actually defaulting to generic theoretical advice.
And when you tell it, give me nuanced, observable things. it's actually gonna go out into its systems and find stuff that's actually proven. The next thing I like to ask it is give me contrarian takes.
Because what I found is that the best performing content on social media often starts with a contrarian take. And so I want to tell my AI systems, hey, go and find these contrarian takes in my content, in other people's content that I could go look at and adopt for myself. And then next up, you want to have the AI system define and redefine terms.
That way you're actually prompting a thought in the viewer. If you redefine the term illustration or you redefine the term, hook, then people are going to lean in.
And what you want to do in your content, if you want to make insanely good content, you want to actually make people think and a lot of times AI will default to the easiest path, right, which is just saying generic shit. But if you tell it define and redefine terms, then it's going to give you a way better output, which brings us to principle number five of your AI content creation journey, which is understanding the difference between a proactive and reactive process.
If you're building your AI systems, and you want to create better content using AI, then you need to adopt a proactive process in your content creation journey. Because if you only have a reactive process, right, where you go into Claude every day and you make a video and you have to restart from zero every single time, then you're not going to be able to scale.
Where on the flip side, if you have a proactive process where AI is constantly telling you things and asking you for things, then you're going to be able to run 10 times faster than everybody else. And what this looks like is adopting AI agents within your content creation process. I'm going to get a little bit deep for everybody watching this and hopefully you stick around because I think this could completely transform how you create content.
using ai and so the first thing you want to do is to install some kind of ai agent whether it's cloud code codex co -work or even grokbot but you want to use some kind of agent so that way you're able to give it a system prompt and say here are your set of instructions that you need to do that you need to use and then it's actually able to run principle one through four everything that i went through in this video you can literally take this video link give it to your grokbot give it to your cloud code system and say run this process for me so that we are not starting from a blank page every single time that you start creating content you're starting from a page that AI systems have already written for you.
And that way you're only giving it feedback on the stuff that it's already created. And so this is what my AI agent looks like today. My exact AI agent that helped me produce this YouTube video or even produce all my short form content.
It pulls my internal and external data from tools like topic finder and you know, my granola, my zoom recordings, and it puts it into a system. It finds the overlap between the stuff that's performing online and the stuff that I've talked about already. Then it actually drafts scripts in my own voice, reviews its own work.
So that way by the time that i see the output it's already half baked and that way it's actually being proactive every single day it reaches out to me and says hey i think you should make videos about this and it's asking me for things so that way i'm never stuck with a blank piece of paper i'm always ready with a next video and then obviously the last step here is i give it feedback i want to make it better and the way that i like to do it is not just give it feedback on this one output but i like to train it on what i believe and if there's any principles that i think it missed i retrain my systems until i believe it's fucking bulletproof and if you want my custom clawed system prompt that I use within my clawed code agent, then just go find me on Instagram and message me the word clawed and I'll send it over to you and watch this video next if you want to learn how to be disgustingly good at content creation.
The Hook

The bait, then the rug-pull.

Most people ask ChatGPT for a viral video and read it off a teleprompter like a robot. This video lays out the five principles used inside a large personal brand to get AI output that still sounds like a person.

Frameworks

Named ideas worth stealing.

01:50model

The 10/80/10 rule

  1. 10% ideation with your data
  2. 80% execution by a frontier model
  3. 10% distilled feedback

Spend human effort at the start and end, and let the model do the bulk of the drafting in between.

03:08list

Four things to train your AI on

  1. Voice rules
  2. Banned words and phrases
  3. Preferred structure
  4. Self-review

Replace long prompts with a standing set of rules the system always applies.

04:55list

Four feedback requests

  1. Novel, non-obvious concepts
  2. Nuanced, observable things, not theory
  3. Contrarian takes
  4. Define and redefine terms

Specific asks to push a first draft away from generic output.

CTA Breakdown

How they asked for the click.

VERBAL ASK
08:40product
“Find me on Instagram and message me the word clawed and I'll send over my custom system prompt.”

Keyword DM offer placed at the very end, right after the proactive-agent demo.

Storyboard

Visual structure at a glance.

open
hookopen00:00
training categories
valuetraining categories02:48
feedback requests
valuefeedback requests06:00
CTA
ctaCTA08:40
Frame Gallery

Visual moments.

One-click upgrade to your Google

Get more breakdowns in your search results

Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.

Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
Watch next

More from this channel + related breakdowns.