Every competitor's ad library tells you exactly what they're afraid of. The trouble is reading it. Most teams scroll Meta Ad Library for an hour, screenshot a few units they like, and call it research. That's not intelligence, that's mood-boarding.
What follows is the four-prompt chain we've been running with 200+ operators across the Onbrand network. It takes a competitor URL, a Meta Ads Library export, and roughly nine minutes. Output: a structured creative-pattern brief that holds up in a Monday strategy meeting.
Prompt 1, Hypothesis. Feed the chain three months of a competitor's live ads and ask the model to cluster them by the implicit promise being made, not the format. You'll typically see two or three dominant promises and one experimental wedge.
Prompt 2, Constraint map. For each cluster, ask: what would this brand never say? The answer is usually more revealing than what they do say. Constraints define brands far more than choices do.
Prompt 3, Eval rubric. Generate a five-axis scoring rubric (clarity, novelty, proof, urgency, specificity) and re-score each ad. The lowest cells are the gaps you can attack.
Prompt 4, Counter-creative. Produce three concept directions that exploit the lowest-scoring axes while staying inside your own brand voice. This is where most teams stop being analysts and start shipping.
Total tokens, end to end: under 14,000. Total time once you've run it twice: nine minutes. The full system prompt and the spreadsheet schema are in the appendix below.
"The point isn't to copy their creative, it's to map the constraint they're optimizing for, then violate it on purpose."