Geo Audit

GEO Audit

Two phases: readiness (can answer engines reach, parse and cite the site?) and visibility (do they actually name it?). scripts/livecheck.js fetches a live URL and checks the infrastructure; scripts/audit.js runs the same checks fully offline on saved files; scripts/cite-suite.js generates the buyer-prompt suite and scores the results you collect.

Hard rules — read before every run

• Report what the engines actually say, not what should happen. Never claim a site "is cited by ChatGPT" from readiness signals alone — only real prompt runs count. Mark readiness findings as readiness, citation findings as measured.
• livecheck.js --url fetches the audited site only (page, robots.txt, llms.txt, sitemap HEAD). No third-party requests. Decline audits of sites the user clearly doesn't own or administer unless it's their own public competitor analysis.
• The prompt suite is a fixed, versioned suite — don't freelance the wording between runs, or month-over-month diffs become meaningless.
• When scoring, every mentioned/cited must come from a real engine response you saw, never assumed. If an engine isn't accessible to you, record it as untested and say so.

Network access (declared — livecheck.js only)

• livecheck.js requests only hosts belonging to the URL passed via --url — GET the page, /robots.txt, /llms.txt, and HEAD the sitemap (default /sitemap.xml, or a Sitemap: URL from robots.txt restricted to the same origin). Redirects are followed, bounded to 5 hops.
• Sends a User-Agent header only; reads no local files, transmits nothing but the HTTP request. audit.js and cite-suite.js are fully offline.

1. Readiness audit

Requires Node 18+, zero dependencies.

node <skill-dir>/scripts/livecheck.js --url https://site.com
# offline mode on saved files:
node <skill-dir>/scripts/audit.js --html page.html --robots robots.txt --llms llms.txt

JSON on stdout; exit 1 when warn/fail present, 2 on fetch/read error.

Report checks[]: http-status, redirect, x-robots-tag, https, robots-reachable, sitemap-reachable, ai-crawler-access, sitemap-declared, title, meta-description, robots-meta, canonical, single-h1, json-ld, faq-schema, question-headings, answer-first, extractable-content, freshness, authorship, llms-txt.

2. Citation suite

node <skill-dir>/scripts/cite-suite.js \
  --brand "Acme" --niche "invoicing software" \
  --competitors "FreshBooks,Wave" --audience "freelancers"

Outputs the prompt suite: 12 canonical buyer prompts × 5 engines (chatgpt, perplexity, gemini, google-ai-overviews, claude).

Run the prompts. For each engine you can reach (web search, your own chat access, or the user running them), record per run:

{
  "engine": "perplexity",
  "promptId": "best-for",
  "mentioned": true,
  "cited": true,
  "position": 2,
  "competitors": ["FreshBooks"]
}

Collect runs into {brand, competitors[], runs:[…]} and score:

node <skill-dir>/scripts/cite-suite.js --score results.json

Returns totals (mentionRate, citationRate, avgPosition, shareOfVoice), byEngine breakdown, competitors mention counts, and a verdict line.

3. Verify before reporting

• ai-crawler-access fail → quote the exact User-agent/Disallow lines from the site's robots.txt in your report.
• Position claims need the actual answer text — paraphrase what each engine said, don't just tally booleans.
• A single run per prompt is a snapshot; note variance when you ran fewer than 3 repetitions on nondeterministic engines.

4. Deliver

GEO AUDIT — <domain> — <date>

READINESS: <pass/warn/fail counts> — top fix first
- [fail] <check id> — what it is → exact fix (quote the robots line / tag to add)

VISIBILITY (n=<runs> across <engines tested>)
- Mention rate: <x>% · Citation rate: <y>% · Avg position: <z> · Share of voice: <s>%
- By engine: <e