--- name: audience-builder description: Research targeting options honestly (interests, lookalikes, size estimates) and tell you when broad beats all of them. Use before a launch or when delivery seems stuck in the wrong crowd. --- # Audience Builder ## Doctrine Targeting is mostly the algorithm's job now: Meta removed exclusion options in 2025, retired interest categories in January 2026, and Andromeda aims by reading the ad. Audience work in 2026 is (a) first-party assets (customer lists, lookalikes) that give the system a seed, and (b) knowing when broad is simply better. An honest audience skill often ends with "go broad and fix the angle instead." ## When to use - Before launch, if you have a customer list worth seeding. - When delivery lands in the wrong crowd ("my ad for young women went to older men"): usually an angle problem wearing a targeting costume. - Needs a Meta Ads MCP (read for research, write only for creating audiences you approve). ## Run it ``` Audience research for my offer: [ONE-LINE OFFER + WHO from my angles] 1. Search interests and behaviors related to my category. For each, give the estimated audience size. 2. If I have a customer list or purchasers: propose a custom audience plus a 1% lookalike, with estimated sizes. Draft only, do not create yet. 3. Then be honest with me: given my budget of [$X/day], would broad targeting likely beat these? Explain in two sentences using what my ad creative already signals about who it is for. 4. Wait for my pick before creating anything. ``` ## Guardrails - Research is read-only; audience creation happens only on explicit approval. - Never stack interests to feel in control. If the recommendation is broad, say broad. - If an ad set was built on since-retired targeting options, flag it: those stopped delivering after January 15, 2026. ## Good output looks like A short table of real options with sizes, one honest recommendation, and no audience created that you did not ask for by name.