AI analysis
Every creative gets a structured reading from an AI model, so you can skim a hundred ads in the time it takes to watch five.
On this page
What AI tagging extracts
Each creative is read once by Claude (an AI model made by Anthropic). The model looks at the image or video thumbnail and the copy and returns:
- Hook: the opening line or idea that grabs attention.
- Angle: the persuasion mechanism behind the ad.
- Visual formats: what is on screen.
- Claims made and a one-line summary.
- Emotional driver: fear, aspiration, curiosity, belonging, urgency, relief, pride, humor or trust.
- Target persona: who the ad seems to speak to.
- Offer type: discount, free trial, bundle, free shipping, subscription, or none.
- Production style: UGC (user-generated style), studio, motion graphics, screen capture, lifestyle photo or AI-generated.
These drive the Library filters Angle, Visual, Emotion, Offer and Production, and the pills on each creative, which link to the filtered Library.
The AI analysis panel
Open a creative's AI analysis tab, or the Ad Details drawer, to see:
- An emotion radar: the nine emotional drivers are each scored 0 to 100, and the six strongest are drawn in a fixed order so the same feeling always sits in the same place.
- A content style: testimonial, facts and stats, storytelling, how-to, problem and solution, offer, lifestyle, comparison, humor or announcement.
- A one-sentence creative targeting line, with "See more" for the longer version.
- A persona: age range and gender.
All of this is a model's reading of the ad, not measured data. The panel says so.
Transcripts
Videos get a timestamped transcript on the Transcript tab. Click a line to jump the video to that moment. Timestamps are accurate to about a second. A video with only music or on-screen text shows "No speech detected". The first sentence of the transcript also feeds the brand's Hooks tab.
Transcription runs on Gemini Flash (a Google AI model).
The winning signal and its limits
Each creative carries a Winning Probability score and a label: likely winner, strong signal, testing or weak. It combines:
- how long the creative has run,
- how many ads reuse it,
- how many copy variants it has,
- whether it is still live,
- Meta's "scaled" flag (the advertiser runs it in 3 or more variants),
- and whether its landing page stays the same.
Its limits: it is an inference, not a result. Meta does not publish spend, impressions or ROAS (return on ad spend) for commercial ads, and adintel does not measure them. Use the score to decide what to look at first, not as proof that an ad is profitable. Meta also does not document exactly when it sets the scaled flag.
Credits used
One credit is one US cent of what the work actually costs.
| Action | Credits |
|---|---|
| AI tag on Claude Opus | about 1.7 |
| AI tag on Claude Haiku | about 0.2 |
| Video transcript | 0.3 |
| Pulling a brand's ads when you track it | 0.4 per ad |
Your Space is charged for the work its own actions start: a tracking pull and the tagging it queues, and uploads of your own ads. The scheduled refresh of followed brands is platform work and is not charged to anyone. Browsing, boards, filters and share links never use credits.
See your balance in the sidebar, on Library status and on Plan and billing. When credits run out, paid actions pause until they refresh on your billing date or you buy more; browsing carries on.