The Wrong Pixel

I show AI doing real work inside ads, funnels and dashboards. Visual breakdowns you can steal for your own growth stack.

[The Wrong Pixel @thewrongpixel]

Just Grok Bot and 560 ChatGPT ads I gave it before going to bed. By morning it had found the headline formula that wins, and you can steal it below.

It started with a question I couldn’t shake. ChatGPT now shows ads right under its answers, and nobody really knows what works in a spot like that.

AirOps keeps a public library of these ads, so I pulled 560 of them from 10 brands. I asked Grok Bot to find what separates the ads that keep showing up from the ones that disappear, and went to sleep.

The first thing I saw in the morning was that every classic copywriting trick had lost:

Question headlines: 3.8 average appearances
“vs” competitor headlines: 2.6
Headlines with “Free”: 4.1
Headlines that start with a verb: 11.1
“ClickUp vs Everyone. We Win.” was shown exactly once, and all 14 competitor-attack ads together collected 53 appearances.
The top ad was “Build a KPI dashboard,” four plain words that appeared 338 times.
The formula behind it is a verb plus the user’s task, written the way they’d say it themselves:

Weak: “Which PM tool will your team use?” (37)
Strong: “Automate your workflows” (55)

It makes sense once you look at where these ads sit, right under an answer the person just asked for. Nobody types “ClickUp or Asana?” into ChatGPT, they type “how do I build a dashboard,” so the ad that reads like the next step of that answer is the one that gets shown.

Here’s the prompt to rewrite your own headline:

“Here’s my ad headline: [X]. Rewrite it 5 ways as verb + the task my user types into ChatGPT. Max 5 words. No questions, no ‘free’, no competitor names.”

[The Wrong Pixel @thewrongpixel]

Stop brainstorming ads. Your competitor already did it for you, with their own budget.

Every ad your competitor has paid to run sits in Meta’s Ad Library, open to anyone, and most marketers scroll through it once, save a few screenshots and never come back.

I wanted to use that library as research, so I had AI sort Speak’s ads into a map and then built a new ad from what the map showed.

Step 1: collect the real ads

We pulled Speak’s ads from Meta’s Ad Library and saved the creative, the copy and the launch date for each one, so everything after this step was built on ads that ran.

Step 2: split the motive from the format

The Researcher sorted every ad along two lines: why someone would want the product, and how the ad makes them want it.

“Speak without freezing” is a buying motive. A creator story, an animation or a product demo is a format. With the two kept apart, you can see which messages keep coming back, how each one gets presented, and which pairings nobody has tried yet.

Step 3: find a connection worth testing

Two ads sat close together on the map. One nailed the pain, the moment you know the words in another language and freeze the second someone speaks to you. The other showed an AI tutor correcting an answer, which showed the product but never touched that feeling.

The Creative Director combined them: someone gets an awkward question at dinner, steps away to rehearse the answer with the AI, and comes back to the table ready to reply. The pain and the fix now happen in one short scene.

The workflow you can reuse

collect real ads → separate motives from formats → find a useful connection → build your next test

You end up with a specific creative idea and the real ads it came from, which gives the AI far more to work with than “write a high-converting ad.”

The library shows what a competitor is running, but it won’t tell you which ads make them money, so treat whatever you build as a test you still have to run.

The prompt

Collect ten to thirty of a competitor’s ads from Meta’s Ad Library, with the creative, the copy and the launch date for each, and give them to Claude with this:

You're a creative strategist. Below are real ads from [competitor],
collected from Meta's Ad Library, each with its copy, a description
or image of the creative, and its launch date.

1. Map the ads on two separate axes:
- Buying motive: why someone would want the product
(e.g. "speak without freezing").
- Format: how the ad delivers it
(e.g. creator story, animation, product demo).
Show the map as a table: motives in rows, formats in columns,
ad examples in each cell.

2. Point out what you see:
- which motives repeat most often
- which formats each motive gets paired with
- which motive and format combinations are missing

3. Find two ads that would be stronger together, for example one
that captures the pain well and one that shows the product well.
Explain why the combination works.

4. Write one new ad concept from that combination: the situation,
the moment the product appears, and how the story ends.
Include the hook, a short script and the CTA.

Don't claim any ad is profitable or performing well. The library
doesn't show results, so treat everything as a hypothesis to test.
My product: [product, audience, what makes it different].
Save it for your next creative sprint, and when you open a competitor’s ad library, start by sorting what they promise from how they show it.

[The Wrong Pixel @thewrongpixel]

@aroido_bigcat Agree. But the key idea is just to make your product routine more interactive and engaging

[The Wrong Pixel @thewrongpixel]

The most boring minute in your product could be its best feature, if you turn it into a game.

Think about the moments in your product when people just sit there: waiting for a file to upload, scrolling a catalog for the fifth minute, or clicking through the same sorting task they did yesterday.

Most teams treat those minutes as dead time. I wanted to see what happens if you turn one of them into a small game, and how fast you can build that with AI before committing to a full product build.

The experiment: picking something to watch

I started with the moment almost everyone recognises, when you open Netflix and spend twenty minutes scrolling without picking anything.

So we built Tonight’s Pick, a small game for choosing what to watch. Posters fly across the screen, you swipe through the ones you don’t care about to slice them, and you tap the ones you like to save them. Once three of them are saved, you pick one and you’re done.

The idea that made it work came during the iterations. The first version had random monsters as targets, and it was fun but pointless. When we swapped the monsters for real posters, every move in the game started building your shortlist, so playing and choosing became the same action.

It’s a browser prototype and a concept, not a Netflix feature, and Netflix is just the example everyone already knows.

Three places to try this

While people wait. Your product is generating a result, loading a file or building a report, and in the meantime the user gets a small scene to play with, maybe one that lets them try out the controls of whatever is being made or shows them a feature they haven’t found yet.

While people choose. Instead of a long catalog, a few quick actions build a list of what they like, and the same idea could be adapted for clothes, trips or design templates.

While people do routine work. Sorting files, clearing an inbox or practising vocabulary gets quick gestures, visible progress and satisfying feedback, and every move in the game is a piece of the real task getting done.

The pattern behind it

Each of these follows the same chain:

moment in the product → simple action → instant reaction → useful result

For our example, that’s choosing a film → slice or keep a poster → the poster splits or fills a slot → a short list of things to watch.

Four rules so the game doesn’t get in the way
The player can skip it at any time.
The result never waits for the game to finish.
Game effects never stand in for real loading progress.
The usual way of doing the task stays available.
How we built it

Grok found game references on X, ChatGPT matched the mechanic to a real user task and wrote the spec, and Claude built the browser prototype and the demo. After a few rounds we had the controls and visuals where we wanted them.

What you get from this is a working prototype you can put in front of users to see whether it helps them get through that step or just distracts them. We haven’t measured the effect on retention or conversion yet, and that’s the next thing to test.

The prompt

If you’ve thought of a boring minute in your own product while reading this, here’s a reusable prompt for finding the right mechanic and building the first version:

Help me find a spot for a short, useful interactive moment in my product.

Product: [description]
User: [who they are]
Current flow: [steps]
Problem moment: [waiting / choosing / routine]
What the user needs to get or do: [result]
Materials and style: [links, images, references]

Suggest three different mechanics using this pattern:
moment → action → feedback → useful result.

For each one, describe:
- what the user controls
- how the action connects to their actual task
- what happens if they skip it
- how the scene ends
- what signal would show it's helping

Pick the simplest one and build a browser prototype that works with
mouse and touch. The demo should fit in 15–25 seconds.

Use real user actions and give instant visual feedback. Never make
someone finish the game to reach the product. If the main result is
ready early, offer to open it right away.

Build the working mechanic first, then style it from the references.

Pick one moment in your product. Build the smallest playable version. Watch whether people finish the task, or just play with it.
Save the prompt for your next prototype.

[The Wrong Pixel @thewrongpixel]

@6vintik9 Giving the landing agent the actual ad helped. It had a specific promise to follow through on, instead of starting from scratch with “write a landing page.”

[The Wrong Pixel @thewrongpixel]

A full marketing team for a $20 product? Easy, and the result is better than most agencies send back.

I took a $20 skincare product from Naturium and handed the work a small marketing team would normally do to three AI agents: one did the research, one wrote the ad and one built the landing page.

I ended up with an ad built around a single reason to buy, plus a page that carries that same promise from the moment someone clicks.

Most brands lose people right at that click. The ad says one thing, the visitor lands on a homepage talking about something else, and now they have to figure out for themselves whether they’re in the right place.

The researcher

I gave the first agent the product page, the photos, the price and the customer reviews, and asked it to find the strongest reason people buy this product.

It came back with who the buyer is, what makes them buy, the facts that back it up and the photos that show it. I asked it to source every claim or mark it as an assumption, so I could see which parts were facts and which were guesses.

It settled on [BUYING MOTIVE], and the ad and the page were built on that.

The creative director

The second agent got the research and was allowed to sell exactly one motive, so it couldn’t fall back on listing every benefit the product has.

It wrote one ad with the copy, the image placement and the button text: [AD HEADLINE].

The landing builder

The third agent got both the research and the ad. A page written from the product alone tends to wander off from whatever the ad promised.

I asked for a page where someone who clicked recognises what they came for straight away, so it opens on the same promise as the ad and keeps it all the way down to the button.

Claude then put all three outputs together into the assets you see above.

Why I didn’t do it in one chat

When I run everything through a single chat, I keep re-explaining the product at every step, and by the time I get to the landing page the original idea has gotten lost in everything said along the way.

With three chats, each one only works on its own piece. I decide what moves forward, so the motive from the research makes it into the ad and the page without getting watered down.

Setting it up

Make one project and put the product brief in it, so every chat works from the same facts.

Open a separate chat for each role. Once you’re happy with a result, paste it into the next chat yourself rather than expecting it to remember.

To test another angle, keep the product facts, swap in a different buying motive and run the ad and the page again. You’ll have a second ad and page pair to test against the first.

[The Wrong Pixel @thewrongpixel]

@VARnaldinho The handoff was the part I wanted to test: could the same buying motive survive all three steps? We got the ad and page built. Whether anyone buys is still an open question.

[The Wrong Pixel @thewrongpixel]

@6vintik9 "AI traders do an 18x in 36 trades" sounds cool until you realize $3.6k is basically just one lucky meme coin gamble that the bot happened to catch on a trend. Still, the pipeline with separated agent roles is pretty slick, can't deny that.

[The Wrong Pixel @thewrongpixel]

PROJECT BRIEF
Product: [name, URL]
Price: [price]
Photos: [attach]
Customer notes: [reviews, DMs, support questions]

CHAT 1: RESEARCHER
Find the strongest reason someone buys this product.
Return: the buyer, the buying motive, supporting facts, usable images.
Rule: every claim needs a source or is labelled as an assumption.

CHAT 2: CREATIVE DIRECTOR
Input: the approved research.
Return: one ad with exact copy, image placement and CTA.
Rule: sell one buying motive only.

CHAT 3: LANDING BUILDER
Input: the approved research and the approved ad.
Return: a landing page that continues the ad's promise.
Rule: the visitor should recognise what they clicked for straight away.

NEW ANGLE
Keep the product facts, replace the buying motive, rerun chats 2 and 3.

[The Wrong Pixel @thewrongpixel]

Paste this before your next landing page (or run it on the one you have):

Act as a blunt performance marketer who has reviewed thousands of landing pages. Audit this page for default AI patterns, in design and copy.
Flag every default: tiny all-caps badge above the headline, gradient text or one colored word in the headline, three identical cards with an icon in a rounded square, emoji as icons, glowing blobs, purple-to-blue everything, “Trusted by 50,000+” or logos I didn’t give you, words like supercharge, seamless, unlock, effortless, elevate, and any headline that could sit on any product.
For each one, one line on what a real visitor feels when they see it.
Rewrite it so only this product could use it:
– headline that names the customer, the result and the timeframe, in words a customer would say out loud
– one-sentence subheadline that says how it works, no adjectives
– show the product doing its job (a real table, chart, schedule, menu) instead of feature cards
– proof I can back up, or [ADD REAL PROOF]. Never invent numbers, logos or testimonials.
– CTA that says what happens after the click
Name the one change that makes this page look least like the other 50 in my category.
Short sentences. No buzzwords.

[The Wrong Pixel @thewrongpixel]

99% of AI landing pages are the same page. Here’s how to build the 1%.

Your landing page looks like the other 99%.

The 1% banned 6 things:
No caps badge over the headline.
No gradient text.
No “Trusted by 50,000+”.
No 3 identical icon cards.
No emoji icons.
No “supercharge”.

Same Claude, but totally different sites.

The full prompt ↓

[The Wrong Pixel @thewrongpixel]

@ryangrowthh thanks, really simple and useful guide. I'm now trying to make big things with AI in marketing and start my journey soon. If you wanna see how AI real works inside ads - coming soon