website-ai-discoverability-audit

name

website-ai-discoverability-audit

description

Audits a public website for search crawlability, structured data, sitemaps, robots.txt, optional llms.txt, and major AI crawler policies; produces an evidence-backed report and only applies vendor-confirmed remediation guidance. Use when the user wants an AI/search discoverability audit, llms.txt review, robots.txt for GPTBot/OAI-SearchBot/Claude/Perplexity, sitemap quality, or “make my site work well with AI search + Google/Bing”.

Website AI & Search Discoverability Audit

Goal

Run a practical, evidence-first audit of a site’s discoverability for:

• Traditional search crawling/indexing (Google/Bing baseline)
• AI-related crawlers and AI-search surfacing (where vendors publish explicit rules)

Then output:

1. A status per checklist item with URLs, HTTP status, and short evidence (what was fetched/parsed).
2. Recommendations for gaps and risks.
3. A “Authoritative fixes” section containing only actions that are directly supported by official vendor/spec documentation (see reference.md).

If something cannot be verified from the public web response or official docs, mark it Unknown and propose how to verify—do not guess implementation details for a specific framework.

Preconditions (must clarify fast)

1. Primary production origin(s) to audit, e.g. https://www.example.com (include www vs apex if both exist).
2. Whether the intent is maximize visibility, maximize privacy/opt-outs, or mixed (common: allow search bots, disallow training crawlers).

If unclear, ask once, then proceed with the best default: audit the user-provided URL exactly as given.

Audit method (how to gather evidence)

Prefer live fetches (same as real crawlers see):

• Fetch with normal browser-like headers where possible.
• Record: final URL after redirects, status code, content-type, obvious block pages (403/401/5xx), and caching headers if relevant.

Minimum fetches (adjust for redirects):

• GET /robots.txt
• GET /sitemap.xml and any Sitemap: targets discovered in robots.txt
• GET /llms.txt (optional signal; absence is not automatically a “failure”)

Spot-check 3–5 representative templates (home + primary commercial page + docs/blog/contact or local equivalents):

• View HTML source for <link rel="canonical">, hreflang, JSON-LD blocks, title/description patterns
• Confirm meaningful body content exists as HTML (not only empty shell requiring client-only rendering for core facts)

If automated parsers are unreliable, still report raw findings and mark parser confidence.

Checklist (what to audit)

Mark each row: Pass / Fail / Partial / Unknown.

A. Robots & crawling policy

• /robots.txt exists at the correct host root and returns 200 with plausible text rules.
• No accidental global disallow for important user agents (especially *).
• Sitemap: directives are present, valid absolute URLs, and reachable.
• AI crawler policy is intentional (not accidental default):
  • OpenAI: OAI-SearchBot, GPTBot, ChatGPT-User (see reference.md)
  • Anthropic: ClaudeBot, Claude-SearchBot, Claude-User
  • Perplexity: PerplexityBot, Perplexity-User
• If using WAF/CDN bot management: flag risk of silently blocking verified crawler IP ranges (Perplexity explicitly calls this out).

B. Sitemaps

• Sitemap(s) are reachable, parseable, and include the URLs that matter for discovery.
• For XML sitemaps: evaluate whether lastmod exists and looks truthful (Bing emphasizes honest lastmod as a freshness signal).
• Note sitemap scale limits per sitemap protocol (split/index when huge).

C. HTML & on-page signals (baseline quality for humans + machines)

• Reasonable <title> + single clear h1 pattern on audited templates.
• Internal links are real <a href> where navigation matters (crawlability).
• Canonicalization strategy is not obviously self-conflicting across duplicates (www/apex, trailing slashes, localized duplicates).

D. Multilingual / multi-region (if applicable)

• hreflang consistency (reciprocal expectations), localized URLs resolve, no accidental noindex on one language variant.

E. Structured data (Google-oriented baseline)

• If the site has content types eligible for rich results, check for JSON-LD presence on representative templates.
• Validate mental model against Google’s structured data policies: mark up only what users can see; avoid empty schema pages.

F. llms.txt (optional curated “LLM map”)

• Fetch /llms.txt if present; if absent, treat as optional unless the user explicitly requires it.
• If present, sanity-check it matches the documented shape from the llms.txt proposal site (H1, blockquote summary, H2 sections with markdown links, optional ## Optional section semantics).

G. AI-search / indexing “freshness accelerators” (optional)

• If Bing ecosystem matters: consider IndexNow as an optional complement to sitemaps (not a replacement).

H. Google quality gates (high-level)

• If content is heavily automated: align messaging with Search Essentials and spam/scaled-content policies (do not promise ranking outcomes).

Output format (required)

Produce a report with these sections:

1. Scope: audited origins, date, and pages/templates sampled.
2. Executive summary: top 5 risks/opportunities in plain language.
3. Findings table (mandatory columns):
   • Area (A–H above)
   • Check
   • Status
   • Evidence (paths/headers/short quotes—keep quotes minimal and non-sensitive)
   • User impact (search visibility, AI surfacing, privacy)
   • Recommendation (what to do next)
4. Authoritative fixes (strict rules):
   • Only include items backed by reference.md.
   • Each bullet must name the exact lever (robots.txt group, bot user-agent, sitemap field, doc URL).
5. Unknowns / follow-ups: explicit verification steps (e.g., Search Console, server/WAF logs, template-level schema validation).

Authoritative fixes you may state confidently (vendor-grounded)

These are safe to include as concrete guidance when relevant to the user’s intent:

• OpenAI: OAI-SearchBot vs GPTBot are independent robots.txt controls; OAI-SearchBot impacts ChatGPT search surfacing; changes may take ~24 hours; ChatGPT-User is user-initiated and is not the right lever for “Search opt-outs” (use OAI-SearchBot per OpenAI docs).
• Anthropic: separate bots for training vs user-directed retrieval vs search optimization; bots honor robots.txt; avoid relying on IP blocking alone as a durable opt-out mechanism; Crawl-delay is supported for throttling (non-standard extension).
• Perplexity: treat PerplexityBot vs Perplexity-User separately; Perplexity documents that the user fetch path generally ignores robots.txt; provide WAF guidance requiring UA + published IP ranges.
• Google: robots.txt must be at site root, UTF-8; Sitemap: directive is a standard discovery mechanism; structured data is for explicit semantics; JSON-LD is generally recommended when choosing a format; follow Search Essentials / spam policies for scaled/automated content.
• Bing / Microsoft AI discovery: treat XML sitemaps + truthful lastmod as important freshness signals; IndexNow is optional and complementary.
• llms.txt: follow the public proposal/spec site for format and intent; do not claim it is a Google ranking requirement.

Do not invent:

• Framework-specific code changes unless the repo is open and the user asked for implementation.
• “AI citation guarantees” or numeric lift claims without a measured experiment in that property.

Progressive enhancement (how the user should evolve this skill)

When repeated audits reveal stable fixes for a stack (Astro/Next/WordPress/etc.), add a short subsection under Stack playbooks in a separate file (optional) — keep SKILL.md lean.

Additional resources

• Official links and specs: reference.md