name: research-audience description: >- Audience research: who the buyers are, in their own words, where they gather and how to reach them — plus named prospect lists when asked. Use when the marketer says "research our audience", "who should we target", "mine our reviews", "why do people buy us or leave", "where does our audience hang out"; or for lists: "find 50 right-fit leads with emails", "find retailers or accounts to pitch", "score these leads / who do we call first", "who's looking for an alternative to X", "tell me when a past champion changes jobs". Reads the brand's own customer data first and delivers a readable report saved to the brand. Never contacts anyone. Not for market size or trends — use research-market; not for writing the messages — use write-outreach. version: 0.8.3

research-audience

Audience research answers, for the people a brand sells to: who they are (buyer groups, sized in buyers, with buying roles), what they want, their problems and how we solve them, their words (the customer-quote bank), where they gather, and how to find and reach them. It owns the brand's one list of buyer groups and the quote bank; market research reads them and sizes the market around them (its figures are context; group sizes are counted here), and every maker reads them. A list of named people is a separate mode, only when asked. The craft is in references/audience-research.md and references/quote-bank.md; this file is the order of work.

The rule that matters most: the brand's own customers first, then free pass, report once (../marketing-guide/research-pattern.md (in the marketing-guide skill) — read it before the first search). One specific question for the brand's customer data is allowed up front; after that, cover every group on every dimension from free sources before the next message. Never stop to ask "which group should I go deeper on?". Paid reads only deepen it, with one credit yes.

Depth matches the ask; keep to these call caps (paid reads are credits, only after the one credit yes):

• Quick ("who's our customer, roughly?", a single question): no workers, about 20 web calls in all, the answer in chat plus a one-page report.
• Standard ("do the audience research", the default): at most three group workers, one level only, each capped (segment-pass § Budget); news done by you (about 6 searches) and sizing gaps the workers left (about 6 more). About 80 web calls in all.
• Deep (the marketer asks for depth, or after the paid top-up yes): up to five workers; the caps stay. Never brief a worker to load this playbook or any skill — the brief is the research-audience-segment-pass skill pasted whole plus the inputs, and nothing else; workers use only web search and page reads, no GooseWorks calls. Sampling and saving the quote bank (references/quote-bank.md) are yours, not the workers'. Workers never start workers. Search before you fetch; fetch only a page you'll quote or count from.

What you get

1. The audience report through the write-report skill (opens with one click, saved to the brand): in short; decisions for you; the buyer groups — ranked, each sized in buyers or companies, who uses / pays / approves, the lead one named — labelled provisional when the evidence doesn't clear the bar; per group what they want, their problems and how we solve them (or "you'd need proof of X first"), their words, where they gather and how to reach them; why they buy and why they leave (from the brand's own data, where it exists); how rivals talk to them; what to do; what we're not doing; what we couldn't check; sources with a source-skew line.
2. Buyer groups and the quote bank — the durable records every other playbook reads.
3. Only in list mode: a list built, a list the marketer brings scored (and inbound leads ranked by who to answer first), or a standing watch — named accounts and people with dated sources, honest contact routes and a buying signal per row (references/list-mode.md).

Before you start

1. Setup first: do the marketing-guide skill §1–2 (connect, brand, brain read) before anything else; the job line is "I research your audience and save what I find through it." Here: call get_started with wants: "more"; when it asks which brand, ask in one line and call it again. On the brain path open the audience and customer chapters, 14 (rules), 22 (what we've made and run) and 23 (what is working), and gooseworks:brand_knowledge_search for customer evidence, synced reviews, pasted exports and earlier research; an audience proposal from an earlier run may still be waiting for a yes. Fallback brand read: gooseworks:brand_read (summary, kit — especially audience — products, competitors, organic, learnings) and gooseworks:knowledge_search for the same. Either path: gooseworks:competitor_read; workspace: research/audience/ (buyer groups, quote bank, last report — a re-run says what changed), research/market/ (the latest market report — its market figures are context; group sizes are counted here), research/competitors/ (rival records and report). Read every outreach/*/results-working.md (write-outreach) with open request lines — Hold needed (exclude those accounts until answered), Find needed (find that role at that account), Quote needed (add the phrase to the bank) — answering each under its line, and the lessons that change the fit. Claims: only the read step of the claims-gate skill now.
2. Since last time. Run memory for this run's job (references/run-memory.md: audience research, prospect list, lead scoring or a watch), including what the ledger shows happened to earlier lists and alerts. Its 3–5 lines open the reply and the report, and this run uses them (weight what replied, don't repeat what didn't).
3. Ask for the brand's own customer data — once, specifically, only when it has customers (audience research only; in list mode it moves to the closing decisions) (paid plans live, reviews of the brand, users or orders mentioned in the brain) and no export is on file. One message, then wait:

   "Before I start: your own customers are the best evidence, and I can't read them publicly. Could you share <the export that fits the company type — references/audience-research.md § Customer data>, plus any cancel reasons, win/loss notes or support themes? Paste or upload it — or say "go" and I'll start from public sources." No customers yet (pre-launch), or the marketer says go: start the free pass. Personal data in an export never goes to the brain; only counts and patterns do.

4. No other question before the report. A list ask ("50 right-fit people with emails", a file to score, "tell me when…") goes straight to list mode — no audience pass first; it uses the saved buyer groups if they exist, else drafts the fit from the ask, the site and the brain, and offers the audience report afterwards. Out of scope (consumers as named individuals, journalists, investors, creators to pay): one line and the nearest thing this playbook does.

Steps

1. Frame — the brand's own side first. If customer data came in, read it first (references/audience-research.md § Customer data): who actually pays (company types, roles, plans, sizes), who stays and who leaves, the reasons they give. Then the brand's own site in the host browser (what it sells, to whom, pricing, logos, its words), the brain's audience text, and the latest market and competitor reports. Draft 2–5 candidate buyer groups by situation — never demographics — each with how to recognise one (usable as ad targeting or list criteria). Where the brand's own facts disagree (site vs brain vs customer data), customer data wins; otherwise make the conflict the first decision.

2. News first. Web search, last 90 days, for what is changing for these people: tools and platforms they must adopt, hiring and layoffs in their roles, budget shifts, regulation, the conversation everyone in the category is having.

3. Research every group. Pick the top three candidate groups for workers (a fourth or fifth only if the evidence demands it, with a light pass from you; its moves are labelled unverified unless they rest on a page you read). One worker per group — a subagent where the host supports it — given the text of the research-audience-segment-pass skill as its whole brief, with the inputs it lists: wants, problems and how we solve them, first-hand words, where they gather, three moves, how rivals talk to them, inside its budget. Launch them together; never a worker per source or per topic. Without subagents, group by group, each written to research/audience/segments/<slug>.md before the next. Never pause for paid reads. When they return, filter: drop any quote from vendor, agency, trade-press, case-study or job-post text (references/quote-bank.md); pick each group's report quotes by the recency rule in references/quote-bank.md § Sampling (older rows stay in the saved bank only); a group without enough recent first-hand quotes shows fewer and says "thin on recent voice".

4. Size and roles. Take the workers' counts; fill only the gaps yourself. For each group, a count of buyers or companies — at least an order of magnitude whenever any proxy exists, labelled rough — (bottom-up: directories, store or company counts, job-post counts, industry figures, the brand's own customer mix), each input sourced or labelled an assumption; community member counts are reach, never group size. Who uses, who pays, who approves.

5. Synthesize — on the evidence you have. Rank the groups on stated criteria (size, pain, fit, reach). Name the top-ranked group as the lead; if it doesn't clear the evidence bar (references/audience-research.md § Evidence bar), label it "provisional — <what's missing>", keep it out of "In short" as a settled finding, and make the evidence gap a decision. Every move rests on sources this run read; a move on a channel it couldn't read is dropped or labelled "unverified — check the community and its rules first", and never becomes "Do this first". For you only where a finding matters ("nothing to do here" is fine). How rivals talk to them; don't follow the herd. Using the market and competitor reports read in setup (don't re-read them), say in one line where this one agrees or differs (a different lead group, a different first move).

6. List mode (only when asked) — runs instead of steps 1–5 ("Since last time" still runs first, for the list's job): follow references/list-mode.md — build a list, score a list they bring, or set up a watch, with the outcomes ledger (the outcomes-ledger skill, entity: person).

7. Write and deliver. Write the report once, in the shape of references/audience-research.md § The report — its headings in its order; "Do this first" names who does it, and every group's unread sources merge into "What we couldn't check", each with what it would change. Check and deliver it through the write-report skill (topic audience; brand copy research/audience/<YYYY-MM-DD>-report.md, keep-history), skipping its "offer to open it" — step 8 is the whole message. Save the buyer groups and the quote bank through the save-to-brain skill (paths below). Propose the