ad-teardown
Skill0 starsChanged 21 days agoMITView on GitHub
Tear down a competitor's ads into a playbook you can act on. Classify every ad's opening into 7 hook types, cluster the ads into themes, read run-length as the proof of what works, find the angles nobody is running, and write counter-briefs plus a "never say" list. Works from ads you collected yourself (ad libraries, screenshots, transcripts).
In this repository6 other files it gives its agents.
- ad-copy-frameworksSkill
- ad-hooksSkill
- ad-library-researchSkill
- campaign-analystSkill
- social-media-specsSkill
- ugc-ad-scriptsSkill
The file
Files
Ad teardown
A teardown answers one question: what is this competitor's playbook, and where is the opening for us? It works on
the competitor's real ads, collected from the public ad libraries (see the ad-library-research skill) or from
screenshots and transcripts the user hands you. Ten to thirty ads from one advertiser is the right size.
Rule zero: describe only what you were shown
The hook a viewer meets is the first two seconds of the video, not the caption. If you have only the copy, say the teardown is copy-only. Never describe a visual, a statistic or an offer that is not in the material. A teardown that invents evidence is worse than none, because it gets acted on.
1. Log every ad in one line
| Field | What to write |
| Format | image, video, carousel |
| Opening | the first line or first frame, quoted |
| Hook type | one of the 7 below |
| Angle | the promise in a few words ("sleep through the night") |
| Offer | discount, trial, bundle, none |
| Proof | review, number, demo, authority, none |
| CTA | the button or spoken ask |
| Running | start date and how long it has run |
2. Classify the openings into 7 hook types
| Type | Looks like |
| Problem-agitate | opens on the pain and makes it worse |
| Social proof | "50,000 people switched", reviews, a creator vouching |
| Question | a direct question the viewer answers in their head |
| Stat shock | a surprising number as the first thing on screen |
| Lifestyle aspiration | the after-state, no problem shown |
| Offer-led | the discount or deal is the opening |
| Curiosity gap | withholds the payoff ("I didn't expect this...") |
Report only the types that appear, each with a count and one quoted example. A brand that runs 80% offer-led openings is telling you it competes on price.
3. Cluster into themes, and weight by run-length
Group the ads by angle. Then sort each group by how long its ads have run. An ad still live after 60 days or more is paying for itself; the advertiser has had every chance to turn it off. The longest-running cluster is their proven winner. New ads that all share an angle are what they are testing now.
4. Find the white space
List 2 to 4 angles that nobody in the set runs. Typical gaps: an audience they never address (men, older buyers, businesses), a use case they ignore, a proof type they never use (a demo, a comparison), an objection they never answer.
5. Write the moves
- Counter-briefs (2 to 3). Each one exploits a specific weakness you saw: a hook type they over-rely on, a claim they cannot make, an audience they ignore. A brief is concrete enough to hand to a designer or a video tool: format, hook (with its mechanic), the line, the proof, the ask.
- Never say (3 to 6 items). Phrases, claims or territory the competitor already owns. Copying their language makes your ad read as theirs. Make each one specific ("'clinically proven sleep'", not "health claims").
Output shape
Competitor: <name> · <n> ads · collected <date> from <sources>
Hook types: problem-agitate 9 ("..."), offer-led 6 ("..."), ...
Themes: 1) <angle>, 4 ads, longest 140 days. 2) ...
Proven winner: <the longest-running ad, quoted, with its run-length>
White space: ...
Counter-briefs: 1) ... 2) ...
Never say: ...
Gaps in this teardown: <what you could not see, e.g. "copy only, no video watched">The last line matters. Say what the teardown could not see, so nobody treats a copy-only read as a full one.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

