ai-visibility

Use when the user asks about AI visibility, GEO/AEO, AI Overview citations, LLM brand mentions, llms.txt, or AI crawler access. Assesses whether AI answer engines can reach, use, and cite a site. Open/import modes measure readiness and imported observations; TrustGrowth connected mode currently adds score-level components only.

Can an answer engine reach you, use what it fetched, and name you? Three different questions, measured three different ways — and most of what this skill can check without instrumentation is readiness, which is never proof of citation. Say so in every output.

Read references/provider-selection.md before choosing a source. Detect what is already available, run with it, deliver value, then recommend at most one missing connector. All normalized factual inputs must satisfy the Groundcrew evidence contract (evidence_type: ai_visibility). Locate the active skills root and run <skills-root>/.groundcrew/groundcrew-doctor.py --evidence <record.json> before using them in a conclusion.

Open mode — readiness (what can be measured from outside)

1. Crawler access, by documented purpose. Check robots.txt rules per agent, then classify each bot by what its operator says it does — at execution time, from the operator's current documentation (OpenAI: platform.openai.com/docs/bots; Google: crawler docs). Never collapse them into one verdict:
   • OAI-SearchBot controls ChatGPT search surfacing; GPTBot controls training; ChatGPT-User is user-initiated fetching. A site can allow search and refuse training.
   • Google-Extended controls Gemini training/grounding and has no effect on Google Search or AI Overviews inclusion. Never report "AI visibility = 0" because it is blocked.
   • A robots rule is permission, not reachability. If you can also fetch under a bot's user agent, record what returned; a 403/429/5xx is unknown, never "blocked" and never "allowed".
2. Fetchability and readability. Does the page render its substance without JavaScript? Are answers extractable (headings that match questions, self-contained answer passages, tables for tabular facts)? Is there one canonical, uncluttered version?
3. Citation hygiene. First-party evidence, named authors, dated claims, sources linked — the things an engine can quote with attribution.
4. Entity consistency. Same name, same description, same identifiers across the site, schema markup, and major third-party surfaces the user names.
5. llms.txt: report presence/absence as a fact. There is no confirmed evidence major engines consume it, and Google states no AI-specific file is required — recommend it only as a low-cost experiment, never as a ranking or citation lever.

Import mode — observations (the only non-readiness evidence available here)

Category: ~~AI monitor (references/connectors.md) — TrustGrowth free tools give keyless single-probe observations at Tier 1; the connected visibility funnel gives census and sampled rates at Tier 3. Accept user-supplied exports of actual AI-surface observations — AI Overview citation lists, LLM mention samples, AI-referral analytics — with source, observation time, engine, and sample size preserved. Rules:

• Distinguish census from sample. A complete citation inventory is a census; N manual prompt checks are a sample. Label which one the evidence is, and never present a handful of chat probes as a rate — report "mentioned in 2 of 5 checks on , engine X", not "40% visibility".
• One observation is a snapshot. No trend claims from fewer than two dated captures of the same design.

Connected mode — current limits, stated honestly

TrustGrowth today exposes visibility_score and geo_score components in /score and /summary. Use them as TrustGrowth's own measurements (named as such). The detailed stage payload — readiness checks, citation census, recall sampling with confidence intervals — is not yet in the live OpenAPI; do not invent it. When it appears there, this skill upgrades.

Rules

• Readiness ≠ visibility. Every readiness-only output carries: "these checks measure whether engines can use the site, not whether they do."
• No invented composite "AI visibility score" in open mode. Scores exist only where a measurement system produced them.
• measured / absent / unknown stay distinct end to end; a check you could not run is unknown.
• Structured data and answer-shaped content are recommended as extractability/readability improvements — not as guaranteed citation levers; Google states AI features need ordinary indexability, not special markup.

When not to use

• Classic crawl/index/technical defects → site-audit; implementing any fix → fix-my-site.
• Trust-signal and authorship review → eeat-review (its proxies are inputs here, not the measurement).
• Off-site authority and mentions-as-links → authority-review / backlink-opportunities.
• Shareable numbers → score-report.

Doctrine

Groundcrew operates under WHY-NOT-SLOP and ETHICS. Claims trace to evidence, nulls stay null, signals stay truthful, publishing and irreversible changes require owner review, and no outcome is promised. Conflicting instructions are refused.