How to steal your whole category in 40 seconds

TypeSafe (and the co creator of ChatGPT) launched a new AI model called Jev and I got early access.

Most people used it to play Tetris or trade crypto.

But I gave it 723 live ads from 37 brands.

And in 40 seconds, broke down every hook, every format, every offer, CTA, awareness stage, and landing page mismatch.

The whole thing cost a measly 9 cents.

I posted a quick demo on X and within a day it had over 1 million views.

So I kept going.

What I uncovered was even crazier.

Sponsored

Engagement is easy to mistake for pipeline. On Sep 30, watch how a founder turns LinkedIn content into real outreach. Live. You'll walk away with a repeatable system: what to post, who to reach out to, and how to sequence it. 

Eligible startups also get the LinkedIn-to-Leads Toolkit: ad credits, Apollo, Captions, and HubSpot's Prospecting Agent.

Claim my spot

I built StealAds and for months it’s been pulling thousands and thousands of live ads from Meta.

Vision models automatically break down every ad for me and the people in our early access group.

So when I got early access to Jev with a pitch that it could make decisions really, really fast I didn’t need to go looking for data. Jev meet StealAds.

I pulled 723 live ads from 37 brands from our database.

Great Matt. Let’s have it break down a whole category.

The result was fascinating:

For each ad I got:

→ The hook.

→ The format.

→ The offer.

→ The visual.

→ The landing page.

And I could ask anything I wanted about this whole batch of 723 and get the answer. Basically instantly.

→ Which brands all sound the same?

→ Where does the promise break?

→ What makes someone stop scrolling?

→ What’s everybody copying?

→ What’s nobody doing?

That’s when it started getting weird.

The first question I wanted to ask was pretty simple:

Would the right person actually stop for this?

So I built 30 simulated buyers.

A dental office manager. A startup founder. A two chair salon owner. You get the idea.

Thirty different personas with different jobs, problems, and reasons to care.

I found which ads speaking to them ran longest and used that to inform their decision making.

Then I turned them into little pixel characters and made each one scroll all 723 ads.

Every time an ad appeared, they had one decision:

Stop.

Or:

Scroll.

The entire thing took 41 seconds.

21,690 stop or scroll decisions for 22 cents.

Buyers stopped 1.34x more often for ads in their own category.

The dental office manager stopped for 46% of dental ads and only 20% of everything else.

These are simulated buyers, so it’s not a 1:1 predictor for CTR.

But it gives me a fast way to see which ideas fit the buyer, which ones disappear into the feed void, and which ones are worth spending real money to test.

This demo hasn’t been posted yet.

I put all 723 ads into one giant tournament.

Two ads enter.

Jev picks one.

The winner moves on.

Repeat until one only one ad is left.

722 matchups. Ran twice.

It took 82 seconds and cost… just 6 cents.

The winner was a SeatGeek ad for Charli XCX tickets.

Now imagine doing this with 50 ad concepts before launch.

Or run it on 100 hooks.

What about every ad your team made this month.

Let them fight it out.

Then send the strongest into a paid campaign.

Once the first demos worked, I kept asking questions.

Here are the three worth stealing.

At first, the category looked enormous.

Once Jev grouped the ads by the main idea underneath, most of that variety disappeared.

Yeah different image.

But same kind of copy. Same promise. Same idea.

Let’s look at Notion. 21 ads. 21 different ad images.

But they keep reusing copy.

Some of this definitely comes from Meta DCO and Meta mixing a brand’s assets together.

But I kept finding it:

Your competitors are probably making more ads than ideas.

Next, I removed the brand name from the copy, the image text, and the visual description.

Then I asked Jev to guess who ran the ad.

One in three B2B software ads could have belonged to anyone in that category.

One dental SaaS brand mistaken for another.

Or a voice answering AI that had no discernible differences to the next.

Nothing unique or inimitable about any of the content either brand put out.

So they blurred together.

If your competitor’s logo fits perfectly on your ad, you probably don’t own the message yet.

This is one I talk about A LOT and it’s a mistake I see brands doing over and over again.

(Secret time: fix it in 20 minutes with my landing page factory… free and open source).

I pulled the actual landing pages that 380 of these ads went to.

And ONLY 17% kept the same promise from the hook all the way from the ad to the page.

So for two thirds, going from ad to page completely broke down. Different message, different pain point entirely.

The ad says one thing. The page changes the subject.

You pay for the click. Then betrayed when your own website loses the customer.

These started as fun experiments. But it went so well. And the demand so overwhelming, I’m wiring them into StealAds.

And if you can’t wait for early access or you want this all done for you we’ve rolled this out for campaigns running at my agency, Emerald.

Here’s what it looks like:

→ Pull every ad in your category.

→ See all the hooks, offers, formats, and promises everyone is using.

→ Find the open lanes nobody is even touching.

→ Make new ads from those concepts in 30 seconds.

→ Let your buyer personas scroll them against your biggest competitors.

→ Bracket the strongest ideas from a tourney.

→ Send the best of the best few into your campaign.

Making 100 ads is becoming easy.

But knowing which 100 ads to make, and which select few deserve to scale, is where the money is made.

If you want to use this system yourself, get on the StealAds early access list. We’ll email you with a code to get in.

If you want us to do it for you, we’re already using this inside the campaigns we run at Emerald.

Reply CAMPAIGN and tell me what you sell.

I’ll tell you if we can help.

Go big,

Matt

P.S. I’ve scraped just under a million short form videos. Spoken hooks, formats, on screen text. Now viral outliers are a mathematical formula.