gooseworks-ai/goose-skills / skills/marketing/research-audience
research-audience
VerifiedAudience 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.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 18.7 KB |
| references/audience-research.md | markdown | 15.9 KB |
| references/contacts-and-compliance.md | markdown | 9.7 KB |
| references/data-sources.md | markdown | 12.3 KB |
| references/fit-and-scoring.md | markdown | 7.6 KB |
| references/list-by-company-type.md | markdown | 4.5 KB |
| references/list-mode.md | markdown | 7.5 KB |
| references/list-sources.md | markdown | 9.9 KB |
| references/output-format.md | markdown | 5.3 KB |
| references/quote-bank.md | markdown | 6.3 KB |
| references/run-memory.md | markdown | 5.2 KB |
| references/score-a-list.md | markdown | 7.6 KB |
| references/signals-and-hooks.md | markdown | 6.1 KB |
| references/watches.md | markdown | 4.7 KB |
| skill.meta.json | config | 6.3 KB |
| tests/test-by-hand.md | markdown | 8.9 KB |
references/quote-bank.md
View on GitHub# The customer-quote bank — how to read buyer evidence and keep it
Load in the main run: at the filter after step 3 and at the save step. Workers get only the short rules in segment-pass. research-audience owns the bank; every maker, market research and competitor research read it. A bank is only as good as who is speaking: one honest buyer beats ten vendor blog lines.
## Who counts as the audience's voice (first-hand)
Counts:
- the brand's own customers and prospects — the customer data the marketer shared, reviews of the brand, support themes, cancel reasons, win/loss notes (`source_class: private` for anything pasted or exported);
- buyers and users of rival and adjacent products, on review sites read on the site itself;
- people in the group describing their own work, in community threads, forums, comments and their own public posts.
Never counts as audience voice (keep as context at most, never as a quote row):
- **vendor marketing** — any company's blog, landing page, case study or "alternatives" page, including a big company describing its own team (a vendor writing about itself is curated, not a buyer);
- **agency copy** written for their clients;
- **trade press and analyst articles** — a journalist's or analyst's summary, or an unnamed "brand rep" quoted in one, is second-hand; use it for context and find the person's own words;
- **job posts** — they are the employer's words about a role; use them for wants and sizing, not as the audience's voice;
- **quotes second-hand** on someone else's page — find the original or leave it out;
- people who aren't plausible buyers (fans of a founder, candidates complaining about recruiters, competitors' staff, resellers).
## Sampling
- Sample inside the run's budget (SKILL § Depth): stop a source when it adds no new theme. A standard run keeps about 25–40 rows across groups; a deep or paid run more.
- Write the bank once, at the save step; quotes are copied from the source, never retyped from memory.
- **Stratify by rating** (1★ failures and objections; 2–3★ trade-offs; 4★ "good but"; 5★ outcomes and identity words) and sort by lowest and newest — review widgets lead with 5★.
- **Recency:** weight the last 18 months; at least 70% of kept quotes from that window. Older rows only for themes still alive, labelled with their year; a theme resting mostly on rows older than 24 months is "dated" and never carries the lead group.
- Mark verified purchases; mark and down-weight incentivised reviews ("received free", gifted); syndicated duplicates count once; the same person across sites counts once.
- Count what you read per source (n read, star mix), not only what you kept.
## Quote integrity
- Verbatim only, typos kept, cuts marked `[...]`, never stitched from two places.
- `capture`: `page` (read on the page), `snippet` (the site's own text in search results), `fetch-excerpt` (a summarising fetch tool's fragment — counts for themes, never as an ad line until seen on the page). Re-open up to 3 kept quotes and say how many matched.
- Every row: link, platform, posted date (or "undated"), retrieved date, rating if any; private rows cite the export or document and its date.
- Strip names and handles of private individuals; keep self-descriptions they wrote ("ops lead at a 40-person agency"); strip health, medication and other sensitive details, emails, phone numbers and order numbers.
- `about`: `brand` · `prospect` · `competitor:<name>` · `category`. Rival complaints are opportunity evidence, not traits of our buyer.
- `contains_claim: yes` on any outcome or efficacy statement — buyer words, never brand proof. `personal_attribute: yes` on anything about health, finances, religion, ethnicity or similar — makers rephrase, never lift. `testimonial_ok: no` by default.
## Tags (one primary type per row)
pain · desired outcome · trigger (why now) · objection / anxiety · alternative tried · habit / inertia · pre-purchase question · specific praise · surprise · price / value · **switch** ("switched from X" to us; "went back to X" — churn evidence) · **why they left** (from cancel reasons and churn notes) · **why we won / lost** (from win/loss notes). Also: the job ("when <situation>, I want to <motivation>, so I can <outcome>"), awareness stage, buying role (user / payer / approver), vividness 1–3 (3 = a line a copywriter would steal).
## Themes
Named in buyers' words. **Established:** 3+ rows from 3+ distinct authors; one platform carrying it → "single source", lower confidence. **Signal:** fewer — shown, never presented as a finding. Product defects go to a separate "for the founder" list.
## Source skew — say it
If more than half of a group's quotes come from one kind of source, name it and how it bends the conclusion: complaint sites over-represent angry customers; developer forums over-represent developers; missing LinkedIn and closed forums hide in-house marketers. A conclusion that rests on a skewed source is labelled so in the report, and never sits in "In short" unqualified.
## The saved bank
`research/audience/voc-bank.md` (title `Voice-of-customer bank — <Brand> — <D>`, tags `[research-audience, voc-bank]`, supersede): opens "Do this first, because… <which quotes to brief from>"; one line on what was read and the source skew; then the themes in buyers' words, each with its quotes — quote, who said it (self-description), where, date, link; claim or personal-condition quotes marked "buyer words only — don't use as a claim". Filterable copy: `voc-bank.jsonl`, one row per quote: `id`, `verbatim`, `about`, `capture`, `platform`, `url`, `posted`, `retrieved`, `stars`, `self_descriptor`, `type`, `theme`, `segment`, `buying_role`, `vividness`, `contains_claim`, `personal_attribute`, `source_class`, `status` (ids stable; a dead link becomes `retired`). Same row schema the makers already filter on. First run with an older `research/market/voc-bank.*` on file: import those rows once, re-checked against the rules above, ids kept.
## Before handing over
Any quote without a link or private citation → fix or retire · a group trait with no quote → evidence it or mark unknown · one platform carrying a whole theme → say so · rival complaints merged into our buyer → split · all 5★ → re-sample · vendor, agency or job-post text in the bank → remove · nothing surprises the founder → look again at 1–3★ and at why-they-left rows.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |