--- name: set-up-audience-signals description: >- Configures the audience signal stack on a Performance Max asset group so the campaign starts learning from the best available first-party data instead of guessing. It inventories usable lists, layers signals in quality order (customer lists and site converters before interest categories), builds intent-based custom segments, and handles search themes with an explicit cannibalization check against live Search keywords. Use it when launching a new PMax campaign or asset group, when an existing asset group is underperforming on thin or interest-only signals, or after new first-party data lands. Signals are steering hints, not delivery limits, if you need hard audience restriction on Video or Demand Gen, that is set-up-audience-targeting; if the customer list itself is missing or stale, run build-customer-match-lists first; for a monthly performance pass over live segments use optimize-audience-performance; for the readiness-gate decision use expand-audience-reach. --- # Configure PMax Audience Signals ## Purpose Performance Max decides its own delivery, but the audience signals you attach determine how fast and how cheaply it finds converters during ramp-up. This skill builds a layered signal stack for one asset group, strongest first-party data first, Google interest categories last, and makes the search-theme decision deliberately instead of by default, because careless themes can pull search traffic away from dedicated Search campaigns. ## When to run - A new PMax campaign or asset group is being launched and needs signals before enable. - An existing asset group is slow to learn or performing poorly and its signals turn out to be thin, generic, or interest-only. - New first-party data became available: a fresh Customer Match list, a new converter segment, a linked channel. - An account review found asset groups with no signals or with themes duplicating Search keywords. ## When NOT to run - The Customer Match list is missing, under ~1,000 matched users, or stale → `build-customer-match-lists` first. - You are configuring Video or Demand Gen audience targeting (real delivery restriction) → `set-up-audience-targeting`. - You are reviewing live audience segment performance and pruning losers → `optimize-audience-performance`. - You are deciding whether the account is ready to expand reach at all → `expand-audience-reach`. ## Prerequisites - The PMax campaign and the specific asset group are identified (one asset group per run, signals should differ by asset-group theme, not be copy-pasted). - Site tagging is live so converter and visitor segments exist and populate (HUMAN STEP if tagging is absent, GTM/site work is outside VigilDog). - A Customer Match list uploaded and matched, ideally 1,000+ users refreshed within 30 days (build it first if not). - A list of the account's top-converting search queries, or access to pull them. ## Procedure 1. **Consult the policy guardrail.** Call `gads_policy_guardrail` before the first Google Ads write of the session and respect any flags on personalized-advertising categories. 2. **Inventory first-party assets.** Call `gads_list_user_lists` and grade each list: a Customer Match list is usable at 1,000+ matched users refreshed within 30 days; site segments (all visitors, converters) are usable from ~100 users, meaningful from 1,000. Record what exists, its size, and its freshness. 3. **Gate on the inventory.** IF no usable Customer Match list exists, either route to `build-customer-match-lists` now or proceed without it and schedule a return trip, but say so explicitly rather than silently launching thin. IF converter/visitor segments are missing entirely, flag as HUMAN STEP (outside VigilDog): remarketing sources and tagging must be fixed before this asset group can get its strongest signals. 4. **Attach the first-party layer.** Call `gads_add_audience_signal` on the asset group with, in this order: the primary Customer Match list (all customers for growth goals, the top-revenue slice for efficiency goals), then the site-converters segment, then the all-visitors segment. Preview first (validate_only default); apply only after explicit user approval. These carry the most learning weight, do not launch without them. 5. **Build intent-based custom segments.** Call `gads_create_custom_audience` twice if inputs allow: - a search-intent segment from 10–15 of the account's highest-converting queries (transactional phrasing, no brand terms, no one-word generics); - optionally a URL-interest segment from 10–15 specific competitor and industry pages (deep URLs, not homepages). Preview → user approval → apply, then attach each via `gads_add_audience_signal` (again preview → approval). 6. **Add a thin interest layer.** Attach one or two of the most specific in-market categories that match the asset group's theme via `gads_add_audience_signal` (preview → approval). Add life-event or affinity categories only when the product genuinely keys off a life change or the goal is awareness, and never as the only signal. Demographic filters only when household income or parental status truly narrows the buyer. 7. **Decide search themes with a conflict check.** Call `gads_list_keywords` across active Search campaigns and compare against the candidate themes. IF the account runs Search campaigns covering those queries, THEN skip themes, PMax will otherwise compete for the same auctions and starve the Search campaign. IF PMax is the only search-eligible campaign, or the user explicitly accepts the tradeoff, THEN call `gads_add_search_themes` with category-level themes (broad phrasing, not long-tail keywords; hard cap 50 per asset group). Preview → approval → apply. Record the rationale either way. 8. **Undo cannibalization when detected.** If a later review shows PMax eating Search impression share or the channel mix skewing after themes went live, call `gads_remove_search_themes` (preview → approval) and note the change for the learning-period clock. 9. **Set exclusions.** HUMAN STEP (outside VigilDog) or route: campaign-level audience exclusions (recent converters on a 7–30 day window; the all-customers list for acquisition-only campaigns) are configured at the campaign, not the asset group. Confirm with the user that they exist before launch, a signal stack without exclusions leaks spend onto people who already bought. 10. **Keep the top layer fresh.** On the ongoing cadence, refresh the Customer Match layer with `gads_add_customer_match_members` (preview → approval) per the schedule owned by `build-customer-match-lists`. 11. **Summarize the stack.** Report per layer: what was attached, size/freshness, theme decision and rationale, exclusion status. Expect 14–21 days of learning before judging the effect of signal changes. ## Decision rules - **Layer order is fixed:** customer lists → site converters → all visitors (→ channel-engaged viewers if available) → intent custom segments → 1–2 in-market categories. Interest-only stacks are a launch blocker, not a variant. - **Usability floors:** Customer Match usable at 1,000+ matched and ≤30 days old; site segments serve from ~100 users but only start informing learning around 1,000. A small real list still beats no list, attach it. - **Custom segment sizing:** 10–15 inputs. Below ~8 the segment is too narrow to model; dumping 50–100 keywords dilutes it into noise. Converting transactional queries only. - **Search themes:** IF active Search campaigns cover the queries THEN no themes. IF no Search overlap THEN category-level themes only, ≤50 per asset group, never duplicating live keywords. Treat themes as an explicit opt-in to PMax competing on search inventory. - **Windows:** converter segments 90–540 days (buying cycles are long); all-visitors 30–90 days (intent decays). - **Aim for 5–7 distinct signal types per asset group,** customized to the asset group's theme, identical stacks across differently-themed asset groups waste the structure. - **Evaluation delay:** any signal change restarts meaningful learning; wait 14–21 days before reading results. - **Done means:** first-party layer attached, at least one intent custom segment live, interest layer thin and theme-matched, the search-theme decision recorded with its keyword-overlap evidence, and campaign-level exclusions confirmed. An asset group missing any of these is not launch-ready. ## Common failure modes - **Interest-only signals** (in-market/affinity/demographics with no first-party layer): the campaign learns from Google's broadest guesses and burns budget during ramp. Always anchor on lists and converters. - **Keyword dumping into custom segments** on the theory that more inputs help. They don't; 10–15 curated converting queries outperform a 100-term pile. - **Adding search themes while Search campaigns run the same queries**, the classic self-inflicted wound: PMax wins the internal auction, Search "declines," and the shift is misread as a Search problem. Run the step-7 keyword check every time. - **Exclusions forgotten** because they live at campaign level while this work happens at asset-group level. Step 9 is a launch gate, not a nice-to-have. - **Copy-pasting one signal stack across every asset group**, each asset group's theme deserves its own custom segments and in-market picks. - **Waiting for a "big enough" customer list.** A refreshed 1,000-user list attached today helps more than a perfect list next quarter. ## Related skills - Run before this: `build-customer-match-lists` when the first-party layer is missing or stale. - Run instead when the surface differs: `set-up-audience-targeting` for Video/Demand Gen delivery restriction. - Run after this: `expand-audience-reach` to decide the next controlled expansion; `optimize-audience-performance` for the recurring segment-level review on Video/DG surfaces.