Skip to content

gooseworks-ai/goose-skills / skills/marketing/improve-search-visibility

improve-search-visibility

Verified

SEO and AI-answer visibility. Use for a site SEO or technical audit with ranked fixes, competitor keyword and topic gaps, a topic cluster, SEO metrics for domains, a programmatic SEO page-set plan, or checking where ChatGPT and Google name the brand. Delivers the requested audit, comparison, map or blueprint with dated evidence and clear data limits. Also covers app store listings and local search. Drafts changes; never edits live. Full pages go to make-content; paid search goes to run-paid-campaigns.

FileKindSize
SKILL.mdskill md11.0 KB
references/app-store-listing.mdmarkdown2.4 KB
references/battlegrounds.mdmarkdown3.8 KB
references/citation-protocol.mdmarkdown8.4 KB
references/entity-and-local.mdmarkdown4.3 KB
references/finished-fix-examples.mdmarkdown8.7 KB
references/foundation-check.mdmarkdown7.5 KB
references/free-pass.mdmarkdown4.7 KB
references/gap-to-fix.mdmarkdown5.5 KB
references/page-map.mdmarkdown4.8 KB
references/prioritising.mdmarkdown2.9 KB
references/question-list.mdmarkdown5.5 KB
references/report-template.mdmarkdown4.6 KB
references/rival-content-map.mdmarkdown4.3 KB
references/run-memory.mdmarkdown15.0 KB
references/seo-data.mdmarkdown6.7 KB
references/seo-jobs.mdmarkdown9.7 KB
skill.meta.jsonconfig6.7 KB
tests/fixture-reports/audit.jsonconfig6.9 KB
tests/fixture-reports/audit.mdmarkdown5.5 KB
tests/fixture-reports/cluster.jsonconfig6.9 KB
tests/fixture-reports/cluster.mdmarkdown4.0 KB
tests/fixture-reports/gaps.jsonconfig5.7 KB
tests/fixture-reports/gaps.mdmarkdown3.8 KB
tests/fixture-reports/metrics-export.jsonconfig3.5 KB
tests/fixture-reports/metrics-export.mdmarkdown1.9 KB
tests/fixture-reports/metrics-missing.jsonconfig3.3 KB
tests/fixture-reports/metrics-missing.mdmarkdown1.8 KB
tests/fixture-reports/metrics-traffic-only.jsonconfig3.9 KB
tests/fixture-reports/metrics-traffic-only.mdmarkdown2.1 KB
tests/fixture-reports/metrics.jsonconfig5.0 KB
tests/fixture-reports/metrics.mdmarkdown3.1 KB
tests/fixture-reports/pseo.jsonconfig9.5 KB
tests/fixture-reports/pseo.mdmarkdown6.5 KB
tests/fixtures/coldfield.jsonconfig21.4 KB
tests/run-fixtures.pyscript7.7 KB
tests/test-by-hand.mdmarkdown9.4 KB

references/citation-protocol.md

View on GitHub
# Citation check protocol

Load at step 1 (free pass) and step 3 (paid top-up), through the atom citation-check.

## Surfaces and how each is observed
| Surface | Default observation route today | Runs |
|---|---|---|
| ChatGPT with search | free: host browser (private window, not signed in); else the human's paste set | 1 per question, 3 for the 3–5 key questions (2 pasted) |
| Google AI Overview | free: host browser, when Google lets it in; else paste set | 1 (1 pasted) |
| Google AI Mode (`google.com/search?udm=50`) | free: host browser, when Google lets it in; else paste set | 1 (1 pasted) |
| Who ranks for each question | free: the host's web search, labelled "web search" | 1 |
| Google's exact page one | paid top-up only: `gooseworks:data_call_provider` provider `scrapecreators`, path `/v1/google/search`, query + `region` (organic results only — no AI Overview, no People-also-ask, no local pack) | 1 |
| Perplexity / Gemini / Copilot | only when the buyer uses them; host browser or human | 3 |

Every report row says how it was observed: **looked up through GooseWorks** (organic), **agent browser**, **run by you**. Never present an organic read as an AI-answer read. Never call an AI model's API with a local key to stand in for the consumer apps (no vendor keys, ever).

**Wait for the answer to finish.** ChatGPT streams: re-read the page until the text stops changing and the sources show (up to about 60 seconds); an answer that never finishes is retried once, then marked "didn't finish" — never scored as "not named". If ChatGPT asks to log in for one question, retry it once in a fresh private window, then mark it.

**What signed-out checks can't show.** A signed-out answer is not what a signed-in buyer sees: memory, past chats, the paid model and personalisation all change answers. Say so in the report's "couldn't check" section, every time. Other engines: check Perplexity signed out when the buyers are technical or AI-native; Claude, Gemini and Copilot need sign-in for their consumer apps, so name them as not checked rather than borrowing anyone's account.

**Where the browser really is.** Record the browser's real location (an answer showing local prices or a local-language result gives it away). If it isn't the target market, label those rows ("read from India, region set to US") — Google rows then count as a first look, and the paid top-up (which sets the region) is the way to the exact read; ChatGPT's location can't be set, so say so.

**Speed.** Open the question straight in the address bar (`https://chatgpt.com/?q=<question>&hints=search`) and run several tabs side by side; the free pass is about 20–40 page reads.

**Agent browser rules:** a private window for Google; chatgpt.com in a private window, not signed in — a Temporary Chat only if ChatGPT forces sign-in and never in the founder's account; record which was used. On any CAPTCHA or bot challenge, stop and hand that surface to the human; never solve it.

**Minimum paste set (offered in the first message only when the host has no browser or Google blocked it; the human runs them):** 5 questions — up to 4 key questions plus at least one brand question ("what is [brand] [category]" or "is [brand] legit"; pre-launch: the name-collision question); ChatGPT twice, AI Overview and AI Mode once each — 20 pastes. For the "AI says something false" entry, the paste set is the brand questions (3–6) instead of the key questions. Give it as plain steps, no jargon:
```
For each question below:
1. Open chatgpt.com in a private window without signing in (use a Temporary Chat only if it insists on sign-in). Paste the question, wait for the answer, click "Sources" if shown (else write "no sources"), then copy everything back here under a line like:  Question 3 · ChatGPT · try 2
2. In the same private window, search Google for the short version. If an "AI Overview" box shows, click "Show more" and copy it with its links under:  Question 3 · AI Overview  (if none shows, write "none").
3. Open google.com/search?udm=50, paste the full question, copy the answer and its links under:  Question 3 · AI Mode
```
Source titles and domains are enough. Pasted rows are scored "k of 2, run by you" and labelled with "settings as you reported".

## Honest sampling
- ChatGPT: new chat per question; memory and custom instructions off; never mention the brand in a non-brand prompt. Record whether it searched (sources shown). An answer with no sources comes from training memory: label it — page fixes will not move it quickly.
- Google: incognito, signed out, country set (`gl`, `hl`), city noted for local. Record whether an AI Overview appeared at all; "no AI Overview" is not a miss.
- AI lists rarely repeat between identical prompts: how often the brand appears is the signal; its rank in one list is noise.

## Per-run record
question id · surface · run # · date and time · country/city · observation route · search used (Y/N) · answer verbatim (saved in the report appendix) · brand named (Y/N) · prominence (recommended first / in a list #k of n / passing mention / absent) · own domain cited (Y/N, URL) · every cited URL with source type · competitors named in order · how the brand is described (exact phrase) · accuracy errors (brand questions) · tone (positive / neutral / negative / caveated).

Source types: our site · competitor's own page (comparison, alternatives, pricing, docs) · listicle / "best X" article · review platform (G2, Capterra, Trustpilot, Google reviews) · Reddit / forum · YouTube · Wikipedia / Wikidata · directory · news / PR · marketplace / app store.

## Scoring
- Per question × surface: **named k of N**; **won** (named in ≥ ⅔ of runs), **partial**, **lost** (never named while competitors are), **no-brand answer** (nobody named — only the citation is at stake). **Unstable** = named in some but not all runs, for any N; a row with one run is never "stable". Stability applies to AI surfaces; a Google organic row is spot-checkable as in/out of the top 10 (position ± 3).
- Mention rate per surface and per stage; own-citation rate; third-party coverage (cited pages that name the brand); share of voice against the brands the engines actually named (the tracked roster often misses them; list new names for research-competitors).
- Google organic: best brand URL position in the top 10, else "not in top 10" (one lookup returns page one only).

## What can and cannot be claimed
Can: "On [date], from [country], logged out, [surface] named you in k of N runs; [rival] in N of N; it cited [sources]."
Cannot: share of all users' answers; how often anyone asks the question in AI apps — only vendor prompt-volume estimates exist; use one only when the brand has that account, label it an estimate, never invent one; a "rank in ChatGPT"; traffic or revenue impact; that a fix caused a later change from one re-run (variance is larger than most effects; call a trend after 2–3 runs).
Always print n runs, date, location and logged-out state on the report.

## Search Console, in plain words (optional ask)
```
If your site is connected to Google Search Console (Google's free dashboard of which searches find you):
1. Open search.google.com/search-console and pick your site.
2. Click "Performance" on the left, set the date to the last 3 months.
3. Click "Export" (top right) → "Download CSV", or just screenshot the Queries and Pages tabs.
Not set up? No problem — I'll note it, and setting it up is one of the easy fixes.
```

## Credits
The paid top-up follows the proxy branch of the `spend-gate` skill (`../../spend-gate/SKILL.md`): **one yes** to a ceiling "up to N credits", labelled an estimate, asked in the first message after the free result — never before it, never twice. N comes from the credits log; with no history, say what the top-up adds ("Google's exact page one for your 14 questions, plus whether Google has your site indexed — about 15 lookups") and let the human name the ceiling; the first lookup (the site-wide `site:` check) is the probe inside that same yes. Stop at the ceiling; log every charge in `research/credits-log/<YYYY-MM>.md`; report credits charged in the gate's words. Never call the price-check tool for these lookups (it prices only media generation, not search lookups). Never show your own sum as a price. Page imports (`knowledge_import` import mode, only when the host cannot fetch pages) fit inside the same ceiling.

## Re-runs
Diff only rows observed the same way both times (route and run count); a route change is reported as "method changed", not as movement.
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall