AI Ad Creatives
4 min read
Your competitor has been running the same ad for 90 days. Nobody keeps paying for an ad that loses money, which makes that ad the cheapest market research available to you, and it's sitting in a public library you can open right now.
The free Claude skill ad-library-teardown breaks competitor ads into hook, angle, format, offer, and proof, counts the patterns, and maps the gaps nobody in your niche is running.
This post covers what gets extracted, why the gap map matters more than the pattern list, and the rule that keeps this research rather than copying.
Key takeaways
Every ad gets broken into 5 components, with counts attached to every pattern claim rather than impressions.
Ads running 60 days or longer get weighted as the ones most likely working, because longevity is the only public performance signal available.
The deliverable is a gap map: an angle by format grid showing what's occupied, what's empty, and what's empty for a reason.
Patterns, never clones. The skill extracts structure and refuses to reproduce competitor copy or brand assets.
What gets extracted per ad
Component | What is captured |
|---|---|
Hook | First line or first visual, verbatim, then classified as data, contrarian, open loop, personal, or before state |
Angle | Which objection the ad answers |
Format | UGC video, static, carousel, founder cam, demo, meme |
Offer | Discount, bundle, guarantee, free shipping, or none |
Proof | Reviews, numbers, press, expert, or demo as proof |
Then it aggregates, and every pattern claim carries its count. Not everyone is doing UGC, but 6 of 9 ads open on a data hook is a fact you can act on. The skill needs at least 5 ads per competitor before it will make pattern claims at all; below that everything gets labelled as a thin sample.
The longevity signal
You can't see anyone's ROAS, but you can see start dates in Meta's Ad Library, which is searchable by brand and needs no login. An ad still running after 60 days is presumed to be working, because the alternative is a competitor voluntarily burning money for 2 months.
So the skill asks you to note start dates and weights the long runners accordingly. A brand new ad tells you what a competitor is trying. A 60 day old ad tells you what survived.
The gap map is the actual deliverable
Patterns tell you what the niche is doing. The gap map tells you what nobody is doing, which is considerably more valuable, since matching your competitors gets you competitor results at competitor CPMs.
The output is an angle by format grid marked with who occupies each cell and the longevity evidence behind it. Cross the empty cells against your own strongest angles and you have a test list.
One honest caveat the skill insists on: an empty cell is not automatically an opportunity. Sometimes nobody runs the price angle in a premium niche because it does not work there. The gap map distinguishes empty from empty for a reason, which stops you from testing into a wall that others already hit.
An empty cell is only useful if you can fill it
Finding 3 unclaimed openings means producing for 3 new formats. Vibelets generates on-brand product ads from your store in minutes, no designer or brief required.
Patterns, never clones
The skill will not reproduce a competitor's copy, brand assets, or trademarked lines into your briefs, and that restriction is deliberate rather than decorative.
Cloning an ad copies the surface and misses the reason it worked, which is usually the fit between an angle and an audience you don't share. Extracting the pattern gives you something that transfers. It also keeps you out of a legal conversation you don't want.
How to set it up in Claude
Download the skill file from the skill's Google Drive folder, free, no email gate, then add it to Claude by uploading it as a skill under Settings or attaching it to a project or chat. Anthropic's Help Center guide to skills covers the upload path.
This one is paste first, because the Ad Library does not export. Screenshots, copied text, or descriptions all work. Note each ad's start date as you go, since the longevity weighting depends on it. Aim for 5 or more ads per competitor.
Conclusion
Your competitors have already paid for the testing. Read the patterns with counts, weight the long runners, then go find the cells nobody has claimed. The gap they left is worth more than the ad they ran.
Turn competitor ads into a test list
Free Claude skill, yours in 1 click.
Frequently asked questions
Is the Meta Ad Library free to use?
Yes. It is publicly searchable by brand and requires no login, which is what makes competitor ad research the cheapest research available to an advertiser.
Why do ads running 60 days or longer matter more?
Longevity is the only public performance signal. Nobody keeps funding an unprofitable ad for 2 months, so a long runner is presumed to be working and gets weighted accordingly in the teardown.
How many competitor ads do I need for this to be useful?
At least 5 per competitor before pattern claims mean anything. Below that the skill labels observations as a thin sample rather than presenting them as patterns.
Does an empty cell in the gap map always mean an opportunity?
No, and the skill says so. Some cells are empty because that combination does not work in your niche. The gap map separates genuinely unclaimed openings from cells that are empty for a reason.
Will this skill copy a competitor's ad copy for me?
No. It extracts patterns and refuses to reproduce competitor copy, brand assets, or trademarked lines. Cloning copies the surface and misses why the ad worked for their audience.
Sources
Make ads that sell
your Products.
Generate on-brand product ads in minutes, no designer or brief required.
Trusted by 400+ growing brands