skills/ads/references/creative-research-automation.md
An agentic workflow for running the creative-strategy research that usually eats most of a strategist's time — ad-library teardowns, review→persona mapping, and organic competitor analysis — as repeatable agent runs instead of monthly manual reports. Adapted from Dara Denney's Claude Cowork practice ($100M+ Meta spend).
The core reframe: don't ask the agent to replace the strategist. Offload the research — the part that's slow, mechanical, and where most hours actually go. The agent opens the browser, reads the pages, scrapes the data, and hands back a structured artifact you steer and use.
This is the paid-social creative research cut. For structured competitor dossiers from a URL list, hand off to competitor-profiling. For deep voice-of-customer analysis and JTBD, hand off to customer-research. Persona output feeds positioning.
Point the agent at a competitor's active paid creative and get back a structured teardown of what they're running and who it's for.
Prompt pattern (fill the brackets, paste the real link):
Do a creative analysis on [brand]. Their Facebook Ad Library is here: [exact ad-library URL]. Open it with the Chrome connector. Report on the schema below. If a field can't be verified from the library, mark it "unknown" — don't guess.
Output schema (one report per brand):
| Field | What to capture |
|---|---|
| Active-ad count | How many ads currently running |
| Product lines | Which products/offers the ads promote |
| Creator partners | Named creators/handles in partnership ads |
| Video/image split | % video vs. % static |
| Video-duration distribution | Buckets (e.g. <15s / 15–30s / 30–60s / 60s+) |
| % partnership ads | Share flagged as paid partnerships |
| Messaging pillars | The 3–6 recurring angles/claims |
| Inferred personas | Who each cluster of ads appears to target |
| Top-10 by impressions | Ranked, with what each leans on |
Useful follow-up in the same chat: "where are these ranking by impressions?" and "which of these have been running longest?" (longest-running ≈ proven winner). The % partnership ads and creator partners fields feed partnership/creator strategy; the format split + duration feeds the format taxonomy an ad brief starts from.
Turn a competitor's (or your own) product reviews into personas grounded in real customer language — and surface the gap between who the creative targets and who actually buys.
Three chained steps, same chat:
The signature move — persona mapping. Ask the agent to compare two things side by side:
The gap is the insight. Creative aimed at a 25-year-old early adopter while reviews are dominated by 45-year-old repeat buyers means the targeting-in-creative is off — a concrete brief for the next round. This is the paid-creative complement to full customer-research; persist the personas doc as shared context for both.
A monthly organic teardown of a competitor's (or an admired brand's) owned social — separate from their paid Ad Library.
Prompt pattern:
Do an organic teardown of [brand] on [platform]: [exact profile URL]. Open it with the Chrome connector. Give me follower count, top reels/posts by likes with direct links, what they're doubling down on, and their strengths + gaps I can exploit.
Output:
Run it against your competitors, your clients' competitors, or brands you admire for inspiration. Ask follow-up questions against the generated report in the same chat. For a full structured competitor dossier (pricing, positioning, SEO), hand the shortlist to competitor-profiling.