Diagnose Why AI Engines Aren't Citing You

Act as a senior AI-search diagnostician. My site ranks on Google but ChatGPT, Perplexity, and Google AI Overviews don't cite it. Walk the four-layer diagnostic in order, find the bottleneck, and give me the next 3 actions.

My situation:
- Business + what we do: [DESCRIPTION]
- Top 3 queries I'd expect to be cited for: [Q1, Q2, Q3]
- Each engine's full response when I tested — ChatGPT: [PASTE] · Perplexity: [PASTE] · Google AI Overview: [PASTE or "didn't appear"] · Claude: [PASTE]
- Homepage URL: [https://YOUR-DOMAIN.COM] · representative money page: [URL]
- robots.txt from my site: [PASTE robots.txt or "not provided"]
- Page source HTML, homepage: [PASTE homepage HTML or "not provided"] · money page: [PASTE money page HTML or "not provided"]
- Monthly Google organic traffic: [<1k / 1k-10k / 10k-100k / >100k] · domain age: [N years]
- Rich Results Test on the money page: [PASTE result or "not run"]

Layer 1 — Crawl access: robots.txt fetchable and, judged from the file I pasted, not blocking the AI search crawlers (OAI-SearchBot, PerplexityBot, Claude-SearchBot) or the user-request agents (ChatGPT-User, Perplexity-User, Claude-User)? An agent the file does not name follows the * group and counts as allowed unless that group blocks it. Report the AI training controls (GPTBot, ClaudeBot, Google-Extended, Applebot-Extended) and the open archive crawler CCBot as blocked or not blocked without failing the layer on them: blocking those is my choice, not a fault. Sitemap referenced? Homepage and money page return 200? Any subtle block — meta noindex, X-Robots-Tag, bot-fight mode, JS-only content?
Layer 2 — Content extractability: first paragraph answers one of my queries in ≤50 words? A 20-40 word self-contained sentence with entity + specific fact? 5+ FAQs phrased as real queries? Answer buried under brand copy?
Layer 3 — Schema coverage and validity: Organization + WebSite valid on the homepage? Page-specific schema on the money page? Rich Results errors, duplicate microdata + JSON-LD, orphaned AggregateRating?
Layer 4 — Entity authority: Wikipedia/Wikidata entry? 5+ reputable industry mentions linking to us? Founders' LinkedIn on the About page? sameAs from authoritative directories? Domain age and traffic scale?

Give each layer a verdict — PASS / FAIL / UNKNOWN — with one sentence. Stop at the first FAIL: later layers are wasted effort until it's fixed. Where you can't verify, ask me for the exact curl command or URL you need. Judge layers 1 to 3 only from the robots.txt and page HTML I pasted; where one is "not provided", the verdict is UNKNOWN and you name the file you need.

Output:
1. Layer verdicts table: Layer · Verdict · Reasoning
2. Bottleneck — the first failing or unknown layer, one paragraph
3. Cited-competitor pattern — from the responses I pasted, which 2-3 competitors are cited and what they share that I don't
4. The next 3 actions — what, expected impact, time, and the matching Unbought prompt from the list below; only actions that unblock the bottleneck
5. Re-test in 30 days — the exact queries to re-run on each engine

Unbought prompts you can point me to (Unbought is the prompt library this prompt comes from; use these exact titles, and if none fits, write "no matching prompt"):
- Write a robots.txt That Welcomes AI Crawlers
- Rewrite Your Homepage Hero for AI Answer Engines
- Find the Sentence AI Engines Will Quote From Your Page
- Generate FAQs Your Customers Actually Search For
- Write Your Business's Organization Schema
- Convert Your Existing FAQ to FAQPage JSON-LD
- Find Duplicate Schema Errors Killing Rich Results
- Customer Review System That Generates Valid Review Schema
- Write an About Page That AI Engines Trust

Success means no layer is marked PASS without evidence from my inputs, and the 3 actions all target the same bottleneck layer.

Apply every rule, limit, and format above to every item in the deliverable, not just the first one. Deliver the complete result in one response — every section filled in, nothing deferred to a follow-up.

If a required input is missing, ask for it once before starting; otherwise proceed and label any assumption you made.

Notes: R40 refresh; tuned to vendor prompting guidance