Big Idea

The argument in one line.

Ads and organic content have merged, so an AI pipeline that feeds a creator's own organic performance, ad results, and reliable attribution data back into a self-scoring system turns that merge into a repeatable, compounding ad-testing engine instead of a one-off hack.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You run paid ads for a client or your own brand and want a systematic way to turn organic content into ad creative.
You're comfortable using AI tools like Claude Code and Gemini to process marketing data, not just generate copy.
You have enough ad spend and content volume that a feedback loop between organic performance and ad performance is worth building.

SKIP IF…

You're not running paid ads yet and have no existing content or ad-account data to feed a system like this.
You're looking for a plug-and-play tool rather than a custom-built data pipeline.

TL;DR

The full version, fast.

Hormozi says paid ads and organic content have merged: the best move is running your best-saved organic posts as ads with little more than a click-to-grab overlay. Ryan Mathews tested it, 77 traditional ads against 8 Hormozi-style ads, and got a 12.43x ROAS winner with a two-thirds lower cost per sale, but about half the new-style ads still flopped. To make the win repeatable, he built a pipeline: organic content and engagement data plus ad performance and Hyros attribution data all feed into Gemini for transcription and annotation, then into Claude Code, which finds the patterns between winners and losers and turns them into a persistent scoring rubric. New ad variations get tested against synthetic customer avatars before launch, and every new ad's results loop back into the same system so it keeps compounding.

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 – 00:29

01 · The stats that kicked off this experiment

77 traditional ads vs. 8 Hormozi-style ads; the best new-style ad hit 12.43x ROAS and cut cost per sale by two-thirds, but over half the batch flopped.

00:29 – 01:16

02 · The old way ads get built

Rigid direct-response formulas (Ogilvy-style) that sound and look like ads, which is exactly why viewers scroll past them.

01:16 – 02:53

03 · Hormozi's Skool clip: content and ads have merged

Hormozi says paid and organic media have completely merged; his team stopped appending CTAs and now just overlays a click-to-grab banner on top-performing content.

02:53 – 03:28

04 · Why saves beat views

Hormozi: look at the content with the most saves, not the most views, and if you don't publish much, just run everything you have as ads.

03:28 – 04:36

05 · The real results: a 12.43x winner, but half the batch flopped

The Hormozi-style ad improved from an 11.4x to a 12.43x return, but roughly half of the follow-up batch of Hormozi-style ads still failed.

04:36 – 05:47

06 · Step 1: pulling organic content and engagement data into Gemini

An API pulls saves, views, comments, followers and the actual video from Instagram and Facebook; Gemini transcribes and annotates each video, including on-screen text and hooks, not just spoken audio.

05:47 – 06:46

07 · Step 2: adding real ad performance and reliable attribution

Meta Ads performance data feeds into the same Gemini pipeline, and Hyros is added for attribution because Meta's own tracking is unreliable.

06:46 – 07:54

08 · Step 3: Claude Code turns the data into a scoring skill

Claude Code finds the patterns between winning and losing ads and builds a persistent 'Claude Skill': how to write hooks that convert, which ad formats produce the highest ROAS, how to structure a script, how to use social proof, and what kills conversion.

07:54 – 09:24

09 · Step 4: generating and testing new ad variations

One filming batch becomes roughly 30 ad variations (10 hooks x 3 bodies); synthetic customer avatars trained on real ICP data give feedback on deal breakers and buying triggers before the ads run.

09:24 – 10:23

10 · Closing the loop, and what's next

New ad results feed back through Gemini and Hyros into Claude Code so the system keeps learning; next up is adding competitor ad and content data, which also works for platforms like YouTube.

10:23 – 10:31

11 · CTA

Ryan offers to build the system for viewers and points to a related video on where to send ad traffic.

Atomic Insights

Lines worth screenshotting.

Ads and organic content have merged: posting the same clip as both content and an ad, with no CTA beyond a text overlay, now often outperforms traditional direct-response ad formulas.
Save counts, not view counts, predict which organic post will convert best as a paid ad.
A platform's own ranking algorithm is already signaling which content it wants promoted, so building ads from that signal is working with the algorithm instead of against it.
In a real test, repurposing organic content into an ad returned 12.43x return on ad spend and cut cost per sale by two-thirds versus the account's traditional ads.
Roughly half of the ads following the same organic-style strategy still flopped, proving the approach alone isn't a guaranteed formula.
Meta and Google's built-in ad attribution is unreliable enough that feeding it into an AI optimization system risks crediting sales to the wrong ad; third-party tracking closes that gap.
An AI system can turn ad and content performance data into a persistent 'skill' encoding what makes hooks convert, which ad formats produce the highest return, and what kills conversion.
Simulating ad performance with synthetic customer avatars built from real ICP data lets a script get feedback on deal breakers and buying triggers before a dollar is spent on media.
A single filming batch can be split into roughly 30 ad variations by pairing multiple hooks with multiple script bodies.
Every new ad's results get fed back into the same pipeline, so the AI system's understanding of what converts keeps compounding rather than resetting with each new batch.
On-screen text and pattern interrupts in the first seconds of a video can matter more than the spoken hook, so a system that only transcribes audio misses real signal.

Takeaway

An AI feedback loop, not a single ad hack, is what made the winning strategy repeatable.

WHAT TO LEARN

Hormozi's claim that ads and content have merged only worked about half the time on its own; turning it into a system that scores its own winners and losers is what made the results compound.

01The stats that kicked off this experiment

A single test found that repurposing organic content into ads produced a 12.43x return on ad spend and cut cost per sale by two-thirds compared to running ads the traditional way.
Even inside the winning strategy, roughly half the new ads still failed, which is the actual reason an AI feedback system got built.

02The old way ads get built

Classic direct-response ad formulas are built to sound and look like ads, which is exactly why viewers have learned to scroll past them.
A rigid ad framework optimized for persuasion can work against you once the audience recognizes the format on sight.

03Hormozi's Skool clip: content and ads have merged

According to Hormozi, paid media and organic content have functionally merged, so treating them as separate disciplines is now the outdated approach.
Adding a tacked-on CTA to repurposed content can hurt performance; a simple 'click to grab it' banner over content that already earned attention converts better than an appended pitch.

04Why saves beat views

Save counts are a stronger predictor of ad-worthy content than view counts, because saves signal genuine intent to return to something later.
If your content volume is low, the fix isn't a smarter selection process, it's posting everything you already made as an ad.

05The real results: a 12.43x winner, but half the batch flopped

A strategy can be directionally correct and still fail on execution half the time; the 12.43x winner and the roughly 50% flop rate came from the same batch of ads.
Repeatable results require a system that learns from both the wins and the losses, not just one lucky ad.

06Step 1: pulling organic content and engagement data into Gemini

Pulling raw engagement metrics (saves, views, comments, followers) alongside the actual video file, not just the numbers, is what lets an AI system learn from content instead of only ranking it.
A transcription-only pipeline misses the point: on-screen text and pattern interrupts in the first seconds of a video can matter more than the spoken hook.

07Step 2: adding real ad performance and reliable attribution

Feeding paid-ad performance data into the same pipeline as organic data lets a system compare what worked as content against what worked as an ad, instead of treating them as separate questions.
Native ad-platform attribution is unreliable enough that a system trained on it can credit the wrong ad for a sale; third-party tracking is what makes the training data trustworthy.

08Step 3: Claude Code turns the data into a scoring skill

An AI system can convert raw win/loss data into a persistent, reusable scoring rubric rather than re-deriving judgment from scratch on every new ad.
The categories worth training a system to score are concrete: hook quality, ad format, script structure, social proof usage, and what kills conversion.

09Step 4: generating and testing new ad variations

One batch of filmed content can be turned into roughly 30 ad variations by mixing multiple hooks with multiple script bodies rather than filming each variation separately.
Simulating audience reactions with AI personas trained on real customer data can surface objections, like deal breakers and buying triggers, before an ad ever spends money.

10Closing the loop, and what's next

Every new ad's results get fed back into the same system, so the model's understanding of what converts compounds over time instead of resetting with each round of testing.
The same feedback-loop approach extends to competitor content and other platforms like YouTube, even without save-count data, because view and comment data alone can still reveal a unique angle worth stealing.

Glossary

Terms worth knowing.

ROAS: Return on ad spend, the revenue a campaign generates for every dollar spent running it.
Cost per sale (CPA): The average amount spent on ads to generate one completed sale.
Attribution: The process of determining which specific ad or touchpoint should get credit for a sale.
Hyros: A third-party ad-tracking platform used here for more accurate sales attribution than native ad-platform reporting like Meta's.
ICP: Ideal Customer Profile, the defined description of the type of customer most likely to buy.
Synthetic audience: AI-simulated customer personas trained on real buyer data, used to predict how a real audience might react to an ad before it runs.

Resources

Things they pointed at.

00:00toolGemini

00:00toolClaude Code

05:47toolHyros

02:06channelAlex Hormozi's Instagram and Facebook ad library

10:23productFunnel Architecture (Ryan Mathews' agency) ↗

Quotables

Lines you could clip.

00:29

“I believe that the future of media between paid and content has completely merged. Like we have reached the singularity of that now.”

bold, quotable claim from Hormozi himself that reframes the whole video→ IG reel cold open↗ Tweet quote

02:54

“If you can make all of your ads look like content, the platform loves you.”

short, punchy thesis line→ TikTok hook↗ Tweet quote

03:36

“When I made this, it was 11.4x return. Now the same ad is 12.43.”

concrete proof number, easy to caption as a stat card→ newsletter pull-quote↗ Tweet quote

06:56

“Now Claude has the metrics from everything, it has the content, and it finds patterns between the winners and the losers.”

clean summary of the whole system in one sentence→ TikTok hook↗ 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.

metaphor

If you just start doing it immediately, you'll be ahead by a year. So I ran 77 ads the traditional way and eight using Alex Ramosi's new strategy. His approach produced one of our best performing ads at a 12 .43x return on ad spend and cut our cost per sale by two thirds.

But more than half the ads following his strategy flopped. But seeing the potential of this, I built an entire AI agent that learns from the data and optimizes itself. So in this video, I'm going to show you what his strategy is, how I automated it, and how you can both.

So why did Hermosi change his strategy? He recently said that ads have changed forever. So if you look at his ad library right now, you're going to see some different types of ads than he's ran before.

So what is this new way to run ads? Well, to start, let's go over the old way to run ads. And then we're going to look at what Hermosi himself says that it's changed.

So guys like David Ogilvie taught direct response principles, you know, rigid formulas and frameworks that sound like ads. They smell like ads. They sound like ads.

They look like ads. And the problem is most importantly, viewers know their ads and they don't want to watch them.

So what happens? They scroll right past them. And we've been running 77 ads following this traditional framework, but Hermosi has something better.

Let's hear what he said. I believe that the future of media between paid and content has completely merged. Like we have reached the singularity of that now.

It's done. That's not going to happen. It's already happened.

It's here. We pump out 450 pieces or whatever it is of content per week. And you know what's really cool is the algorithm will just tell you which of these are the best and most interesting.

And so then what we do is we take those things and then just run them as ads. And I used to think, because we did this earlier, probably a year ago, we were adding CTAs on the end. So it'd be like, I'd take a piece of content and then I'd be like, hey, by the way, get the blah, blah, blah.

We don't even do that anymore. just take the ad or take the content and then just just literally put an overlay just put a banner that's just like click to to grab a thing like it people get it they're they're like if the thing was good they're like that was good i'll i'll take the action the bad platform you know allows me to take right now whatever the whatever the thing is but like if i'm talking about like more in -depth work stuff like that middle of funnel bottom of funnel stuff the stuff that has lots of saves crush And the reason I think that saves thing is important for you guys is look at the content that you have that gives you the most follows, not the most views.

Look at the content that has the most saves. Try running those as your ads. At the very least, because some of you guys aren't putting out 450 pieces of content a week anyways, right?

And so like if you're putting out 20, just put all 20 up and you'll be amazed at how much better they're performing. And I think, I mean, this sounds obvious now, but like it's 100 % aligned with the platform objectives. Like a perfect platform would have no ads.

And so if you can make all of your ads look like content, the platform loves you. And you guys, if you just start doing it immediately, you'll be ahead by a year. So you heard it from him that now ads that don't look like ads are typically converting better.

And the algorithm is telling you what will convert. And again, we ran seven ads this way, and this actually was outdated. When I made this, it was 11 .4x return.

Now the same ad is 12 .43. And it's also crushing for Hermosi. Okay, now what's interesting again is we ran eight ads this way.

And actually we launched another six or so earlier this week. And most of them flopped. Like I would say about half of them flopped.

in that first batch. And this got me thinking because ultimately like I agree with the strategy, right? Like these things have merged organic content that's working is likely to work on ads, but there's obviously patterns in between what crosses over and what doesn't.

And I wanted to create a feedback loop to where I could train an AI system. on what not only is working on instagram but also what's producing the return right if i can show ai hey this one's producing a 12 .43 x return this one you know hasn't gotten any sales it can learn from those patterns and start to optimize itself so that's what i want to show you right now how i optimized this entire process all right so the first part of this process is pulling over what hermosi's talking about right the organic content from instagram and facebook right we pull this over via an api and i'm not going to get super technical in this video i just want to show you what's possible but This pulls over the data, right?

So how many saves it got, views, comments, followers, et cetera, pulls over all of that. And then it pulls over the actual content, right? And I'll show you why that's important in just a second.

But we get all of that data from Instagram and Facebook. And again, we're taking some of those pieces of content, the best performing ones, and we're putting them straight into meta ads. But where this gets interesting is its ability to actually create new content for you based on what's working, okay?

So it takes all of this, it feeds that into Gemini, and Gemini transcribes those videos and annotates the videos. content and on -screen text. So if you imagine like a reel, it's not just what's spoken.

There's things going on the screen. It might be like a pattern interrupt in the beginning of the reel. There might be like a caption.

in the beginning of a reel if you just use something that can only transcribe you're not going to get all of that data and arguably sometimes what's on the screen as the hook is more important than the words sometimes the words don't even start for five seconds right so that on -screen hook is super super important and gemini is amazing at doing that right so we feed all this data from the api into a gemini api and gemini spits out the transcripts as well as match with all of the data match from instagram and facebook and it feeds us into cloud code but the other piece of this that is just important if not more important is again we're taking some of this content and we're running as ads and we're also running 77 ads the way most people run them and some of these ads are working quite well so we also feed in all of the data from meta meta ads okay so this gets all of the the metrics on What's the CPMs?

What's the CPC? What's all these, you know, different ad metrics down to really granular details. And it also is feeding that same content into Gemini.

So it does the exact same thing. So again, that feeds into cloud code and now cloud knows, okay, what's working on organic, but also what's working on ads, right? And what organic content that's being ran as ads is working on ads and what organic content that's being ran as ads.

isn't working as ads but there's another piece of this that is so important and that's connecting with something that has really really good attribution right so we connect with hyros hyros and the reason we do this is historically meta and google and these different platforms their attribution is really bad which what that means is that if you feed in this data into cloud code a plot isn't going to have great data it might be attributing sales to an ad that the ad didn't actually produce whereas hyros has really bulletproof tracking so we get really really good data on you know what is that return so when you see a 12 .43x return we could be confident that that actually is the return rise with meta you always have to take it with a grain of salt now claude has the metrics from everything it has the content and it finds patterns between the winners and the losers and the ones in the middle right what's the difference what's creating these outliers and the more data you feed in here right it just continues to get better and better and better and then claude can create a scoring rubric right so claude can start to score content that we generate in

the future based on what's worked in the past right so it has now a real data like if you just gave an ad to claude to score it could do it right but now it's based on real data so its rubric is actually really really refined and what it does is it takes all of these patterns between winners and losers and it starts learning and creating what's called a skill that has an understanding of what creates outliers, right?

So it's learning things with very, very specific details on things like how to write hooks that convert, which ad formats produce the highest return that's been, right? For this particular client, they have some skits. They have some, you know, one person on the screen.

They have some raw videos. They have some more high production videos. It's learning all of that, right?

And it's not just which one's... get the most views or the cheapest CPC? No, which ones produce the highest return?

That's all ultimately we care about. And how to structure the body of the script, how to use social proof effectively, right? Because sometimes we use too much and as we use too little, sometimes certain social proof doesn't work quite well.

And then also what kills conversions and what to avoid. So it's learning all of that stuff. And then what we can do is when we come in and we want to generate new ads, for one, we can just say, hey, make me 30 more variations based on our winners.

Right. And it's going to make us variations based on our winners. We can also come in and we can say, hey, we wrote this script.

Right. We wrote this script here. Can you please revise it?

And we feed that into our system, which has this entire skill set built into it, as well as it has what is called synthetic audiences trained on how real customers are responding to predict ad performance. OK, so Claude can build these synthetic audiences based on the real data and also data on our ICP based on real data of not just prospects, but real good customers.

So we can feed these ads into it and it's taking all of this information and then it's getting feedback. from the avatars where they're talking about their deal breakers their buying triggers their scam detector it's in another 10 questions that it asks them and it's taking all of that for 10 different avatars that are based on icp and then it's feeding that back into clod and we're getting a new script written with that and then we get our ad variations which typically we start with something like nine variations which is three bodies and three hooks each or recently we started doing 10 hooks because claude pointed out there's just such a massive outlier with how good the hooks are so now it gives us 10 hooks and three bodies each so you have like 30 ad variations from one batch of filming and then of course this is where it gets super super interesting this now gets ran as we can run this as content we've done that yeah but we can run this as content and it gets fed into

meta ads. And then what happens is this gets fed all the way back through. So that goes back into Gemini.

It gets transcribed. We get the data from Hyros, right? Cloud code gets all of that and it starts learning and it knows I wrote this ad.

I thought it was going to convert because of X. How did it actually convert? Okay, maybe it didn't convert as well as I wanted it to, or maybe it absolutely crushed.

Why did it crush? So this system is just super, super exciting. Where this also gets even more exciting is one thing we're going to add is feeding in competitor ads and content.

So we can go out and we can find content. You can't get save data, but you can get view data and comment data from competitors per client that is absolutely crushing. And we can feed that data in, right?

Both from competitor ads and competitor content and feed that into Gemini. So that it can also learn from other people. And it might not weight this as heavy.

Because obviously if you have your own data. That could typically be better. But where this could be really good.

Is finding unique angles. Going out. You could do the exact same thing for YouTube.

Find unique angles. Find outliers. And you know recommend them to me.

So I think if you're running ads. And you want to increase your return on ad spend. This year I think this is the best way to do it.

For this particular client. I'm super excited to see where this gets us in six months. When it has that much training inside of it.

So if you're interested in just seeing if we can do this for you. you, I'll put my calendar link down below. But also, if you want to know where to actually send your ad traffic to, check out this video on Alex Ramosi's advice for sales funnels.

The Hook

The bait, then the rug-pull.

Ryan Mathews opens with a head-to-head test most marketers haven't run yet: 77 ads built the traditional way against 8 built Alex Hormozi's new way. The Hormozi-style batch produced a 12.43x ROAS winner and cut cost per sale by two-thirds, but more than half of it flopped, which is what sent him building an AI agent to find out why.

Frameworks

Named ideas worth stealing.

01:17list

Hormozi's Content-Ad Merge Strategy

Treat the algorithm's own signal (saves) as the ad-testing filter, not views
Run high-save organic content directly as ads instead of writing new ad creative
Drop the appended CTA; a simple 'click to grab it' overlay is enough
If you don't publish much, just run everything you already made as ads

Alex Hormozi's public claim that paid and organic media have merged, and the tactical changes his team made to their own ad production as a result.

Steal forany brand or agency deciding what to test next as an ad without commissioning new creative

07:23concept

The Claude Skill Rubric

How to write hooks that convert
Which ad formats produce the highest ROAS
How to structure the body of the ad script
How to use social proof effectively
What kills conversion (and what to avoid)

A persistent, evolving scoring rubric Claude Code builds from real winner/loser ad data rather than general copywriting best practices.

Steal forany paid-ads or content team that wants an AI reviewer trained on their own account's data instead of generic advice

04:36model

The Data Pipeline

Instagram + Facebook organic data and video (API)
Meta Ads performance data (API)
Hyros attribution data
Gemini: transcribes and annotates video + on-screen text
Claude Code: finds winner/loser patterns, builds the scoring skill
New ad variations, tested against synthetic avatars, then run as ads
Results loop back into Gemini and Hyros

The full architecture connecting organic content, paid performance, and attribution data into one self-improving loop.

Steal foranyone wiring multiple data sources into an AI agent for a testing feedback loop, not just ads

CTA Breakdown

How they asked for the click.

VERBAL ASK

10:23link

“I'll put my calendar link down below.”

Soft consulting pitch delivered only after establishing credibility through his own case-study numbers; the link itself (https://funnelarchitecture.com, from the video description) is never spoken on camera.

MENTIONED ON CAMERA

10:23productFunnel Architecture (Ryan Mathews' agency) ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.