16:34Kallaway · Tutorial
How to Create Irresistible Hooks (and blow up your content)
A 16-minute breakdown of the curiosity psychology behind viral hooks, built around one three-step formula and five tactical amplifiers.
December 19th 2024The argument in one line.
Cold AI hook generation is generic because it optimizes for average language, not your niche — anchoring it to your own top-performing video data turns a guessing tool into a hook writer calibrated to what actually works for your audience.
Read if. Skip if.
READ IF YOU ARE…
- You create video content on YouTube, TikTok, or Instagram and your AI-written hooks feel generic even when you prompt carefully.
- You have at least 10 published videos and want to reverse-engineer what made your best performers work.
- You are willing to use Claude Desktop and pay for a Sandcastles pro subscription to run the data pipeline.
- You want a repeatable system for any new topic rather than a one-off list of hook templates.
SKIP IF…
- You have fewer than 10 published videos — there is not enough data to build a meaningful rubric.
- You want a free, tool-agnostic solution — the workflow requires Sandcastles credits and Claude Desktop.
- You are looking for a quick list of hook formulas rather than a process for building your own system.
The full version, fast.
The Hook Machine is a four-step Claude workflow that uses the Sandcastles MCP to pull deep analysis of your top-performing videos, sort winners from losers by natural view-count breakpoint, and extract the psychological and grammatical patterns that made each one work. Those patterns become a custom grading rubric. You then feed any new topic into Claude and receive hook options modeled on your proven formats, filtered by topic compatibility, with letter-grade feedback on your own drafts. The closer the data — your own backlog, one competitor at a time — the sharper the output.
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 – 00:54
01 · Cold open / credential setup
HOOKS graphic open; host establishes authority: billions of views, 1M+ followers attributed to hook craft.
00:54 – 02:00
02 · Why AI hook writing does not work
Cold LLMs are not optimized for social data. Copy-paste templates fail because psychology does not transfer across niches.
02:00 – 03:17
03 · Hook Machine overview
Four-step system: build format library, create grading rubric, generate hooks, grade and rewrite drafts. Requires Sandcastles pro + MCP setup.
03:17 – 26:22
04 · Full live tutorial
Screen-share: paste prompt into Claude Cowork, feed 4 channels (40 videos), sort winners/losers, extract per-channel then cross-channel principles, build rubric, generate hooks by topic, grade and rewrite a manually written hook draft.
26:22 – 27:36
05 · Summary + CTA
Recap, free Claude prompt and skill file in description, emoji CTA in comments.
Lines worth screenshotting.
- Cold AI hook writing fails because LLMs are not trained on short-form social performance data — they optimize for language quality, not audience retention.
- Copying viral hooks from another niche usually fails because hook psychology is niche-specific: the same grammar that works for a faceless analytics channel will read as off in a mindset channel.
- The 80/20 of your channel is real: a small cluster of videos drives most growth, and those are the only ones worth extracting hook principles from.
- Paid and brand-deal videos must be excluded from rubric training — under 2% engagement on a high-view video signals artificial spend and will corrupt your data.
- Specificity beats abstraction across every channel: naming a concrete outcome, number, or constraint consistently outperforms vague superlatives.
- The two-sentence amplifier structure — state the claim, then add a contrarian reframe or comparison in sentence two — appears as a cross-channel winner.
- Confident declaratives outperform hedge questions: winners say things are, losers ask whether things might be.
- Throat-clearing openers — any delay before delivering the value proposition — underperform consistently regardless of creator style.
- Running the Hook Machine on a single competitor produces sharper principles than blending four channels, which dilutes toward the average.
- The hook grader is more valuable than the hook generator: line-by-line breakdown on your own draft surfaces exactly where your instincts diverge from the data.
- Format compatibility matters: a conceptual psychology topic should not be forced into a case-study stat template even when that template has a strong track record.
- Analyzing your own 100-video backlog solo — without competitor data — is the most reliable version of the system.
Why your AI-written hooks keep sounding generic.
WHAT TO LEARN
AI hook writing fails not because AI is bad at writing, but because it has no calibration data for what works in your specific niche and voice.
- A cold language model optimizes for plausible-sounding language, not the psychological triggers that make your specific audience stop scrolling — without your channel performance data as an anchor, every hook it writes is a guess.
- Copying hooks from another creator's viral video often fails even when the format looks identical, because the underlying psychology and cultural context are niche-specific and do not transfer without adaptation.
- Sorting your own top videos into a winner/loser split by natural view-count breakpoint is a more honest rubric than ranking by views alone — videos boosted by paid spend inflate the winner pool with false signals.
- The principles extracted per-channel — psychology patterns, trigger words, grammatical structure, anti-patterns — are more actionable than any universal hook formula because they are calibrated to what your particular audience responds to.
- Cross-channel synthesis reveals universal principles only when channels are genuinely comparable; mixing a faceless stats channel with a mindset talking-head produces averaged-out principles that may not serve either well.
- A hook grader that explains its reasoning line by line is more useful than one that outputs a score, because it makes visible the gap between your instinct and the data — which is exactly the gap a repeatable system needs to close.
- Format compatibility screening — rejecting hook templates that do not match the topic type — prevents the most common AI copy-paste failure before it happens rather than after you have already published.
Terms worth knowing.
- Hook Machine
- A four-step Claude workflow that builds a personalized hook-grading rubric from channel performance data, generates new hooks against that rubric, and rewrites drafts with letter-grade feedback.
- Sandcastles
- A social media research tool providing deep analysis of video transcripts, hook formats, storytelling structure, and engagement metrics. Connects to Claude via an MCP plugin.
- MCP (Model Context Protocol)
- A plugin standard that lets Claude connect to external tools and data sources. The Sandcastles MCP gives Claude access to 24 tools including video analysis and channel management.
- Winner/loser line
- The natural view-count breakpoint in a channel's top 10 videos that separates genuine breakout performers from average ones. Only videos above this line are used to build the hook rubric.
- Two-sentence amplifier structure
- A hook pattern in which sentence one states the core claim and sentence two adds a comparison, contrarian reframe, or magnitude amplifier — identified as a cross-channel winning pattern.
- Cowork
- Anthropic's Claude desktop application, required to run MCP plugins and agentic multi-step workflows. The Hook Machine cannot run in the browser version of Claude.
Things they pointed at.
Lines you could clip.
01:08
“When you write hooks with AI, the LLMs just are not optimized for short form social data. Unless you train them extensively, they just do not understand how to write persuasively for social media.”
01:30
“When you do these crazy copy paste from one niche all the way over to the other, they usually do not work either because the psychology does not translate.”
12:33
“When Callaway names a specific outcome, the video wins. When he leans on episode branding alone, it sinks 50%.”
15:24
“Winners say things are. Losers ask questions.”
19:53
“The more competitors you add in one analysis, the muddier it gets.”
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.
analogy
Today we're talking about hooks. If you want your videos to perform better, the most valuable thing you could do is build a hook writing system. This is a framework and process for writing killer hooks every single time.
Now look, I've created some of the most popular hook trainings on the Internet. I've literally studied thousands of the top performing videos, and I've personally done billions of views and racked up over a million followers just because my hook writing was so dialed. But the problem is getting all that hook knowledge into one AI workflow that actually works for everyone has been pretty tough.
The good news is I think I finally figured it out. I'm gonna give it away for free in this video. I call it the hook machine.
It's an AI workflow that combines data and advanced storytelling principles to help you write winning hooks every single time. So in this video, I'm gonna walk through how the hook machine works, why it works so well, and the exact step for step workflow that you can use to implement it today. Alright.
Let's dive in. Now before I do anything, let me just explain why AI hook writing doesn't usually work. It's actually very simple.
When you write hooks with AI, the LLMs just aren't optimized for short form social data. Unless you train them extensively, they just don't understand how to write persuasively for social media.
This is why when you dump in your topic into a cold AI chat and say, write me hooks, it ends up horrible. They always just end up sounding super generic and never compelling. Now most people have accepted that, but instead they try the copy paste method, where they take a super viral hook from a creator all the way in this niche, and they try to copy paste it and apply it to their niche.
But here's the funny thing. When you do these crazy copy paste from one niche all the way over to the other, they usually don't work either because the psychology doesn't translate. The format crossover doesn't work well.
I mean, even the grammar usually isn't working when you copy paste it. And yet all these people try these straight copy paste hook clones because all the fake gurus told them to. So what is the real way to use AI, but actually solve the hook problem?
The solve, in my opinion, is the hook machine. The hook machine is my four step process that combines data with storytelling principles to give you the best chance at writing an original hook that actually works every time. By the way, I'm also just giving you the prompt and the skill file linked below.
It's free. So you can have that. You can grab that, rip it and clawed, and you'll be good to go.
Now I wanna go through a live example and show you exactly how I use this and it coming to life. So the very first step before you can make this work, you need to have a Sandcastle account. Without the MCP, you can't pipe the social data in.
We can't deep analyze the videos. We can't run the data loop. Everything kinda breaks.
So step number one, you have to have a Sandcastle's account. There's a link below if you wanna join. I'm gonna assume moving forward that you do have at least a Sandcastle's pro account.
Alright. The next step is that we need to set up the Sandcastle's MCP. You have the Sandcastle's account and you have Claude, but we have to build a pipe connecting them so that we can automatically tell Sandcastle to do stuff from Claude.
Now if you've already set up the Sandcastle's MCP because you only have to do it once, go ahead and skip to the next section. If you haven't set it up, I put together a video that I've linked in the description, help.sandcastle.ai/mcp. It takes two minutes to set it up.
Very easy to follow the instructions, but you need to make sure you follow that and set this up before you move forward. Go ahead and pause this video and get the MCP set up, and then come back and keep watching. Alright.
So at this point in the video, I'm gonna assume everybody has a Sandcastles at least pro subscription and has the MCP set up. Now we're good to go. This is gonna be sick.
Alright. So now we're in Claude, and we're gonna go through a live demo of the hook machine. This is like my favorite workflow I've ever built.
So what we're gonna do is you're gonna copy paste in the prompt that I have linked below in the description. I'm gonna copy paste that in as well. I'm gonna use Cowork.
What you see is the desktop app for Claude. If you're not on the desktop app, go to claud.com/download. Download the desktop app.
You need to be on the desktop app in order to run Cowork or Code. So here we go. We've got this prompt.
We're gonna paste it in and let it rip. Okay. So what you're gonna see if you if you're not really familiar with Claude, it's gonna build in the progress tab up in the upper right.
It's gonna build the flow in a minute. But first, it's gonna tee this up. So it says, I'm the hook machine by Callaway.
Here's what I'm gonna do. You're gonna give me social channels, and I'm gonna do three things for you. I'm gonna build a hook format library.
This is a bonus. This is like it's just gonna list all the top performing hooks with the Mad Lib for you. You can export and save that out.
This is a bonus. It's gonna create that personalized grading rubric, which is how it evaluates good versus bad hooks.
Then it's gonna build a hook generation engine where you can drop in any topic and it will write new hooks for you. So what we're gonna do is just to keep it quick, we're gonna do the top 10 videos from me as well as three other competitors. So we're gonna do grab the top 10 videos from these channels.
I'm gonna say ignore any videos about personal life that are not related to the topic.
And I'm just doing this because I don't wanna waste any credits or grab any hooks that are like about a a personal vlog that's like not on topic for me. Okay. So first, I'm gonna grab my own channel.
So we've got that here, and I'm just gonna paste it in right here. Then I'm gonna grab three competitors. I'm gonna do Peter Visuals.
This guy is faceless. I don't know who this is, but pretty impressive. I know he watches my stuff, so he's probably gonna see this.
Bonus footage, new guy, growing fast. I like this guy's vibe. And then we're gonna do Briar, the homie.
Now all these are different styles. Right? So when you do hook analysis, you may wanna run this flow just on you alone because that's really the way to get the best data is, like, just you, but you can also run it with other people as well.
And that's why we segment the analysis out across the different people because Briar style is a little bit different. The bonus footage is a little bit different than me. So it's like the hook's rhythm may not be what I want, but we'll see.
So here we go. This is gonna analyze 40 videos in total, 10 per channel times four channels, which is what I said. That's 40 credits, roughly ten minutes of processing time.
Now if you already have some of these processed on the back end, which I do, it'll tell you on the next step, oh, we didn't actually have to run 40 credits. It was all good because you already spent the credits before. So it's not gonna double charge you.
One last thing, and this is really important. Do you want me to include all the videos or screen out any brand deals? Typically, any video with below a 2% engagement rate that's a high performer has paid spend behind it that will artificially inflate the video and give you a false sense of what will work.
We like to screen these out by default. Are you okay with that? Of course, am.
I wrote the rules. Yes. Screen out brand deals.
And then just to confirm, you said ignore any videos about personal life. The topic here, da da da, is that right? And it'll say, yes.
Let's let it rip. So now what it's gonna do is it's gonna go, and it's gonna find, do you have those channels in your watch list in Sandcastles? If you don't, it's gonna go find the channels in the Sandcastles database, add them to your watch list so it can run this.
Any channel you give, we can do this for. There's gonna be no lag. It may give you a notice that we don't have them in the database, so we need to live add them, scrape their whole backlog, import them, and then run this process for you.
It will let you know that if it needs to. I know that all four of these people are in there, and so you can see what it's doing is it's looking. Do I need to add any of these people to the watch list?
It likely won't. All four channels are already in your watch list, now pulling top videos from each. So this sandcastle's MCP is magic.
I've mentioned this before already. If you've used Appify to try to, like, lightweight connect your social media, you see all those people on TikTok that are like, I built a social media marketing engine using Clods.
Like, what they really said is they connected via Apify, and they were able to scrape the metrics and, the captions. And it's actually very surface level. It's good for, like, ranking the videos that are doing the best, which we can obviously do for free, but you're not getting any deep analysis.
So what this does is all that stuff that you're paying Apify for, we give it to you for free, unlimited for free. The scraping, the metrics, the captions, all for free. What we charge credits for is the deep analysis.
We're doing advanced processing with the transcript, the idea, the topic, the hook format, the storytelling format, the visual layout, all the metrics. Everything we're doing advanced as much as you could do, that's a credit per video. And so if you've set up Apify, this will blow your mind what this can do.
So as I mentioned in the upper right, the progress, the way Claude works is it starts to build this kind of, like, project plan for you. So you can see it's already added the four channels to the watch list. Now it's pulling the top 10 videos per channel.
It's gonna make sure all the videos are qualifiers. Right? So if any are about personal background or any of them are off topic, it's gonna screen those out.
If any of paid brand deals, it'll screen those out. Then it's gonna go and sort the top 10 by person. So it's not gonna mash all 40 together.
It's gonna go list of 10. We wanna keep them segmented. As I mentioned, the hook styles are kind of different per person.
So we're gonna keep them segmented, and then it will delineate a line. The ones above that line are winners. The one below those lines are losers.
And so it's gonna try to figure out, like, where are the winning hooks? Within the top 10 videos, is there a breakpoint that we find, like, super winners?
And it's gonna go through that process and then the rest. Now the cool thing about the SandCastle MCP is that it has access to all these different tools. So I believe there's 24 different tools that are baked into the MCP.
So right here, can see it's hitting the channel recap. We also have the analyze tool, which it'll be using. That's the tool that we actually run the analysis via Sandcastles to Claude.
We also have the credit tool. We have the hooks tool. We have a format tool.
There's there's so many tools. Right? So it's giving us an update now that says the Callaway Marketing 10, the Peter Visuals, Briar 10 have been analyzed.
The bonus footage is the only one with gaps, so it's completing those now. So you can see it's firing the analysis tool right here. It's concurrently analyzing all six of the remaining bonus footage videos, so it already has 34 done.
If you already have the videos analyzed on the Sandcastle's web app side, this will be extremely quickly. Any ones that has to fill the gap for, it'll take about one to two minutes per video, and it's running those concurrently. So this should take about another minute.
One other tip for anyone doing this, if the analyze process takes a long time because you're pulling a bunch of videos that have not been analyzed, you can always go into the Sandcastle's web app, and you can manually analyze or bulk analyze the videos you want. It's faster in the web app, and then you can export that data to a CSV and drop that in here, which will actually save you some time on this first analysis piece.
So you kinda have two ways of doing it. The analysis is just faster in the web app than it would be through the data pipe because it has to keep going back and forth. So now we're moving to the next step, which is sorting the winners and losers.
You can see in the upper right, we've got three done. We're now sorting. So if we go down, here's the top 10 that it's hooked from me, and it's identified some as winners and some as losers.
And the reason or the way it did that was it noted a natural break point in the views between video four and five. And so what it's saying is number one, two, three, four, those are big bangers that we should analyze the hook for. The rest were not breakouts relative to the others.
So it did that for my videos, and then for every other channel it does it for, it's gonna break it out as well. So for Peter visuals, there are five with a clear breakpoint. For bonus footage, there are four.
Although he's a bit smaller, so you can see, like, four or five, these are these are kind of similar. So you can change the delineation point in a second, but for what it did, it did the top four. And then for Briar, it also pulled four.
There's a clear breakpoint. So content is just like anything else. It's eighty twenty.
Right? So, typically, there's gonna be a few videos that are huge bangers that drive a majority of the growth. Those are the ones that we wanna study and extract the principles out of.
So if we scroll down, now it's asking us, okay. It did the summary. It screened out paid videos, which, like, Peter visuals, GammaPost, 0% engagement, clear clear paid videos.
So it screened that out because we don't wanna evaluate that hook or value it too highly if it's actually a mid performer that was just boosted. So now it's saying, do you wanna adjust the winner loser line? I'm gonna keep what it did because all of them except for maybe bonus footage look good.
So we'll say keep the line. I'll move to the next piece. So now that it's keeping the line, what it's gonna do is it's gonna analyze all the ones above the line, and it's gonna try to find the core principles for why those hooks did well.
It's gonna first do them separately, one for me, one for Briar, one for bonus footage, one for Peter visuals, and then it's gonna see are there any core principles or pillars that are the same across all of us. Those are a great sign to combine with kind of the universal principles. So that's what it's doing now is it's trying to evaluate all of these.
Now it already has the transcript, the hook. It already did all the analysis. So now it's just running the meta analysis on top of the data that we have in here.
So here we go. Now it's starting to run through the analysis for each person. So what it's doing is it's looking for a couple different buckets.
It's looking for psychology takeaways, trigger words and framing takeaways, grammatical takeaways, and then non obvious takeaways. It's gonna do that for each person, and then it's gonna combine comprehensive insights across.
So for me, for example, these were the four it looked at and said, these are the six, the kind of losers. For on the psychology front, it says, winners attach a concrete outcome to a constrained input. For example, thirty days of content in thirty minutes, losers lean on series branding without a standalone value promise inside the first sentence.
So it's like, okay. On the psychology front, winners versus losers. Trigger words versus framing.
Winners use actually, every day, crushing right now, easiest. Losers are using vague superlatives. On the grammatical structure, winners are declarative and front load the payoff.
That's actually a really interesting insight. Here's how to here's how to do x and y versus the losers, which are kind of delaying the takeaway, and they have no identity.
So here's a pattern. When Callaway names a specific outcome, the video wins. When he leans on episode branding alone, it sinks 50%.
Like, that's a very insightful takeaway just on mine alone, and this only analyzed four videos. Imagine if you just did your own and you ripped 50 videos and you had it run all 50. The amount of rigor in the analysis would be pretty insane.
Now let's look at Peter visuals. It took these five. The psychology, Peter's entire framework is case study as content.
The hook is the stat, but not all stats win equally. Winners pair the stat with a second sentence amplifier, like a comparison or contrarian reframe.
The losers state the stat and stop. So, like, right away, I'm like, okay. That's pretty clear.
Every single one of his videos is this case study format, but it's working super well when he adds this kind of amplifier on the top or on the second sentence. Trigger words and framing, he uses magnitude language and light the light counts as additional proof points.
The exact numbers that he used matter. 90 to one twenty million outperforms 19 to 40, so bigger evidence is better. That's pretty intuitive, but, like, clear that it confirmed it.
Then it goes through the grammatical structure. Here's an anti pattern. Audio clip openers where the video opens with reference videos clip instead of his framing, these underperform by roughly 60%.
Again, very insightful, not obvious, and this is not the type of thing you would get from scraping off the top. You need the actual transcript and the the visual analysis to be able to get this deep. So it does that for him.
I'm not gonna go through all these just to save you guys time, but bonus footage, it's gonna do the same thing here. Briar is gonna do the same thing here for him. You could do this for every single competitor that you have.
Now it's gonna move to the cross channel synthesis. We didn't read all these, but I bet if you read every single one of these, you'd be like, oh, okay. It's kind of common that across these three people, they seem to do the same thing psychologically.
Right? So now what it's doing is it's gonna look for any common attributes across all four. And what it's assuming is all of these people are close competitors, and so you actually want some similarity.
Like, you wanna find if there is a a similar attribute that you can use. Here we go. It says, after running the channels independently, here are the patterns that show up in two or more of the channels.
First is specificity beats abstraction. All four channels show in spec when you're specific, it works better. Good insight.
Number two, the two sentence amplifier structure appears both in Peter visuals and me and somewhat Briar. So it's like you state something and then you give an amplifier in the second sentence. I tend to do that a lot.
Peter does that a lot. It's noticing that commonality. Number three, avoid throat clearing openers.
Briar plus bonus footage plus Callaway all show this. You're not like wasting time or delaying the value. Pretty interesting.
And then here's a few others. Confident declaratives beat hedge questions. Briar Callaway, Peter all confirm.
Winners say things are. Losers ask questions. So, again, these are all, like, really compelling and interesting insights that I would not have known right off the rip.
I can kind of gauge my own stuff, but comparing it to others, this is what an LLM is really designed for. So there's a couple others. Here's some divergences where we do different things.
Okay. Cool. So now what we did is we're taking the universal principles.
These are the eight principles that I believe to be true when it comes to hook storytelling that just crush every time. If you've watched a lot of my videos, you'll recognize most of these. These are common attributes that tend to be true.
Then we're adding on the custom principles that we've extracted from this data. Now if we would have pulled 50 videos from each of these channels, all four of these channels, maybe this would have slightly changed. If we would have pulled other channels, this would have definitely changed.
If we would have pulled just my channels, this would have changed. So you can run this a bunch of times to get different variations of your hook rubric, the grader, and you're gonna use that to write different hooks. So this can be a one and done if you analyze a bunch of videos, but it doesn't have to be a one and done is my point.
And so look at how many kind of universal principles we're adding just from this. Okay. Cool.
So that's the rubric. Eight principles, 11 custom principles with sources tagged. If anything looks off, we say so.
If not, we say let's continue. So I'm gonna say continue. Alright.
So now we're continuing from here. And so first, what it's gonna do is it's gonna build a library of all the winning hooks. This is just like a bonus takeaway.
Like, you don't even need this for the hook machine, but I just included it because if you wanna summarize all the ones you just analyzed via the credits, it has that here. Right? So for me, it's like, here's one video, the original hook, the format.
Second video, original hook format. And you just copy this. You could drop this into a different chat, save this as a skill.
You could export this, upload this somewhere else. Right? So you've got all of the hooks that you just analyzed right here.
Mine, Peter visuals, bonus footage, Briar, they're all here. That's just a bonus, like I said. Now where we're at here is it's time to write a topic.
And so the whole point of this is that you wanna build a hook machine that you can just write any topic, and it will give you winning hooks. And it will use some of the data, some of these proven formats, and it will write its own using those principles. So let's just do a hook example.
So I'm gonna say, I wanna make a video about this concept, dopamine ladders.
This explains the six levels of dopamine that get released when someone watches a video.
This is how you make people addicted to content. Let's just say, like, this is a topic that I wanna make a video about.
This is a unique topic. I validated. I'm like, I wanna make a video about this.
Boom. So I add that topic in there. Now you can add a lot more research if you wanted just for the brevity of this and, like, to show the simplistic workflow.
I'm like, okay. I'm just gonna drop, like, a couple thoughts, couple sentences. So now it's gonna write me hooks about this idea of dopamine ladders.
What it's gonna do is it's gonna give me up to five hooks per person. So five for me, five for Briar, five for bonus footage, and five from Peter visuals based on the winning hook formats. Now it's not gonna do five if there's only two that won.
Right? So it's only it's gonna go up to five based on the ones that actually analyzed. That's why more videos is better.
It's gonna give me all of those. Now it's not just gonna copy paste the format. It's gonna take the format and reskin it using our principles to write it close, but not exactly.
It's also gonna only pick the winning formats that match this type of topic. So it's intuitive enough to know, well, this is kind of like a theoretical psychology concept. So if there's a random hook format that's about stats or a case study, it's not gonna be a match, so we shouldn't even use it.
So it's gonna rule out some of those as well. So it's gonna give you a list based on each of the channels, and then it's gonna write its own list of five that I find are typically the best. So usually, it's like your own channel.
It's gonna give good ones. The originals are gonna be good ones, and then it's hit or miss depending on the channels that you actually pulled because, like I said, Briar bonus footage, these guys yap, and so they're more long winded. And I tend to be a little bit tighter, and I don't wanna have these, like, ramble y hooks, not in, a bad way.
They just talk with more words per sentence. So those might not be a fit for me. So let's just see what it comes up with.
So here we go. Right away, it's saying heads up, Peter Visual's library is mostly case study stat. Dopamine ladders is a conceptual framework with no external stat.
So four of his five formats fail at the compatibility screening. So it's not gonna give me examples. And I actually like that because forcing a case study hook template into a psychology breakdown is actually exactly what I was saying in the beginning of this video.
You wanna avoid this crossover format cross pollination just does not work. So it's intuitive enough to know that, which I think is pretty powerful. So first, it says, okay.
For my own videos, only two of the four fit. So here are the example hooks it wrote.
Here's how to release six levels of dopamine in sixty seconds of video, or dopamine driven content is outperforming everything else, so here's the easiest framework for triggering six layers of it. I kinda like the second one. I don't really love the first one, so let's keep going.
This is the one Peter visuals format that actually worked. One level of dopamine is bad for attention. Three levels is good, and six levels is what makes people addicted.
That that might actually work. Bonus footage. I studied 200 of the most addicting videos on the Internet, but a comment said you're overthinking it.
And the truth is they all hit the same six dopamine triggers I now use in every script. I think there could be something here that's a bit long winded. If you force me to make your next video addictive in six steps, this is exactly what I would do.
This is this is decent. That's actually decent. How to turn any boring topic into an addictive video because dopamine driven hooks are on fire right now.
I'd probably skip this. Now one thing I didn't mention is this is grading these hooks based on the criteria in the rubric.
Right? So it's rating these a's. It's not gonna pull ones that are bad fits.
I've coded it, so it's like, yo. If it's a c or below, don't even give it to me. And so that's why you're seeing these grades.
This one from Briar, if you held a gun to my head and said you had to make a video addictive in the next twenty four hours, I'd basically give you my six level dopamine ladder. This could work for a yapper. It's a little long for me.
For dopamine hooks, single trigger is d tier. Nobody actually stays for one trigger anymore. I don't love that.
I have bad news for all of you copy hook templates. Single trigger hooks are dead in 2026 because dopamine has gone multilayered, and I don't really love that either.
So there's only a couple in this list that I really like. Let's see the originals. There are six levels of dopamine your brain hits while watching a video, and most creators only trigger one of them.
That's pretty good. Most creators trigger one type of dopamine in a video. The viral ones trigger six.
That's pretty good. Some videos release six types of dopamine in under a minute. That's why they feel impossible to stop watching.
There's something good there as well. Addictive videos don't have one good hook. They have a dopamine ladder, and they're exactly six rungs.
I don't love exactly six rungs. Six levels of dopamine separate the videos but separate the videos people watch from the videos people can't stop watching. That's like a good tweet.
I don't know if that's good for a hook. So the point is what we have here now is based on what I put in the topic, we generated hook options for each of these people plus originals. Now as you can see, the more videos that you analyze, the deeper this goes.
And so you actually want to have a lot more videos in the hopper. Right? So here, we only are getting two, three, three.
We're only getting, like, seven well, there's one more. Nine or so that fit because of the given topic that we have. If you analyze your whole backlog and then you've got a ton to pull from, you'll have so many more options.
So this is kind of like the initial list of hooks. Now as a backup, where I think this closes the loop for sure is around this original rewrite.
So what I can now do is I can say, you know what? I don't really like any of these, but this this one gave me an interesting idea. I'm gonna go play around on my side and, like, write something that's a derivative of this.
I'm gonna paste it in, and then I want this to grade my hook against the criteria and also rewrite it for me. So I'm gonna paste in.
Let's just let's just come up with one now. So I'll say, this is how you get people actually addicted to your content.
It's called dopamine ladders. And once you learn it, you'll never look at content the same way again.
Let's just say this is something I came up with. So I'm gonna say go. Now what it's gonna do is it's gonna grade this, and I I bet this gonna give a b minus or a c plus.
It's a little long winded, and it's not specific enough. But let's see. It's gonna grade it, and it's gonna explain why it gave me the grade that it gave me.
Here we go. Gave me a b. So it said line by line breakdown.
This is how you get people actually addicted to your content. It covered this. The distillation was weak.
This is how you is soft. No specificity. Like I was saying, credibility anchored, none.
So it was pretty weak on the first line. It's called dopamine ladders, once you learn it, you never look at content the same way again. Good, but it's vague, and it's missing a concrete asset.
Why not hire? Here it is. Then it's like, okay.
Here's three versions of a rewrite that could improve it. Number one, this is how you get people actually addicted to your content. It's called dopamine ladders, and they're exactly six rungs.
Why it's stronger? Because this is specific. Number two, most creators trigger one type of dopamine in their videos.
The addictive ones use a six level ladder. Oh, that's pretty good. That's my favorite so far right there.
I studied 200 of the most addictive videos online, and they all do the same thing. Climb six rungs of a dopamine ladder. There's something interesting there, but, uh, this is the best right here for sure.
I I would use that one in my video. So the point is, like, if you run this workflow and you just use your own videos, right, that's that's the real like, if I'm if I'm being honest, the real way to do this is I make a 100 videos on my own. I analyze all 100.
I pipe all 100 into this. And you could just if you wanna avoid the Sandcastle's credits, you still wanna use the MCP, but you could just drop in the transcripts of the 100. That could also kind of work.
Although we do a lot of processing visually, so you kind of want the Sandcastle's analysis. You analyze your 100, then it runs the top 20. It grabs the there's gonna be a breakpoint probably for 10 or 12 or 15.
You're gonna give it a topic, and you're likely gonna have a lot of other versions, other videos that map to that topic. It's gonna write you hooks inspired by that remixed here and originals, but the originals are only gonna be based on my storytelling principles and your account, not the competitors. Interestingly, the more competitors you add in one analysis, the muddier it gets.
So you might wanna run this four times with separate single competitor each time and have, like, the Briar hook writer, the Bonus Footage hook writer, the Callaway hook writer, and you go to them separately. Because I have a feeling that when you combine all the insights, it does dilute to the average.
But when you do it one at a time, it makes it super clear. What I love though is this I can put in my own hook, and it will grade it against the criteria that we built from the data and then rewrite it to tweak. The rewrites are like my favorite part of this.
So once you have this done, you don't have to keep running the entire flow again. Like, I could just drop in another topic here. Right?
So it's like, throw me a different topic. So once you have this in here, what you can do is you can actually set up a project in Cowork with this and then just hit that project over and over, or you could just do this in the same thread. This will burn a bunch of tokens though if you, like, have to keep running the whole thing.
So multiple ways to do it, but, like, this this is the best way I found to use data, but also originality and writing to get the AI machine anchored to make hooks that actually win. Hopefully, you guys got a lot out of that. I'm gonna do a bunch more workflows like this where it's in-depth, and I'm kind of, like, more more like raw, like I'm on a podcast kind of riffing, less, like, performative.
So if you guys liked me on this and you liked seeing this walkthrough, please, like, let me know in the comments. Leave the give me the give me the mustache chef in there and that, you know, we're cooking.
So that's that's what I wanna see. Alright, guys. That's all I've got for this video.
I tried a little something new here where I went more in-depth with the workflow, really shared my screen, like, showed literally everything I'm doing. If you like that and you want me to make more of these content meets AI workflows, drop a little chef emoji. I'm happy to do more of these less theory, more tactics.
Just let me know if you like this or not. As always, I'm trying my absolute best to push the envelope and give you guys stuff that you don't typically see. My goal is to give you the stuff that most people put behind a paywall that I'll just give you for free.
You've been around here for a while, you know my goal is to help a million people generate over a million dollars with content. That's a trillion dollars in market impact. That's kinda what I'm going for.
In order to do that, I gotta give these in-depth things for free. So please drop in the comments what else I can cover that would help you unblock or move forward with content. If you're a business owner or creator and you're trying to grow with content, you can probably tell this is gonna be one of the best channels for you.
Please make sure to subscribe, and I've got a ton of free stuff in the description. Make sure to check that out. For this video in particular, I've got the free Claude prompt for the hook machine, the Claude skill for the hook machine, as well as a bunch of other resources.
So make sure to grab that, make sure to use it, and we'll do everything we can to get you guys growing faster. Alright. That's all I've got for this video.
We will see you guys on the next one. Peace.
The Hook
The bait, then the rug-pull.
The word HOOKS fills the frame in white block letters before the host says a single word — a visual commitment that this video delivers, not teases. What follows is a 27-minute live demo of a workflow most creators would put behind a paywall.
Named ideas worth stealing.
02:00model
The Hook Machine
- Step 1: Feed channel URLs into Sandcastles MCP via Claude
- Step 2: Sort top 10 videos per channel into winners/losers by natural view breakpoint
- Step 3: Extract per-channel principles then cross-channel synthesis
- Step 4: Generate hooks filtered by format compatibility; grade and rewrite your own drafts
A four-step AI workflow turning your own top-performing video data into a personalized hook rubric and generation engine.
Steal forAny content creator who wants a systematic alternative to cold AI hook generation
12:00list
Four Hook Analysis Buckets
- Psychology patterns
- Trigger words and framing
- Grammatical structure
- Non-obvious anti-patterns
The four dimensions the Hook Machine analyzes for each channel to distinguish winning from losing hooks.
Steal forManual hook analysis on any creator's backlog
14:54list
Four Cross-Channel Universal Principles
- Specificity beats abstraction
- Two-sentence amplifier structure
- Avoid throat-clearing openers
- Confident declaratives beat hedge questions
Patterns that appeared as winners across two or more of the four analyzed channels, suggesting near-universal applicability.
Steal forA starting rubric before running your own channel data
How they asked for the click.
VERBAL ASK
26:22link
“I got the free Claude prompt for the hook machine, the Claude skill for the hook machine, as well as a bunch of other resources. Make sure to grab that.”
Soft, value-first. Gives away the actual tool before asking for anything. Sandcastles subscription is the monetization layer, not the explicit CTA.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
- Use Sandcastles to grow 10x faster on social media (try free) ↗
- Download Claude Desktop App (so you can use Cowork or Code) ↗
- Sandcastles MCP Plugin For Claude ↗
- Hook Machine Prompt ↗
- Hook Machine Claude Skill File ↗
- lu.ma ↗
- Hire my short-form video editors ↗
- kallaway.co ↗
- Hire My YouTube Strategist (Ideas, Titles, Thumbnails) ↗
- Hire My Thumbnail Designer ↗
- Instagram ↗
- TikTok ↗
- YouTube ↗
- LinkedIn ↗
- X ↗
- Threads ↗
- Instagram ↗
- YouTube ↗
- TikTok ↗
Visual structure at a glance.
