How to get cited by ChatGPT, Perplexity & Claude

How AI citation actually works

Traditional SEO gets your page into a ranked list. AI citation is different: an LLM reads your page, decides whether a passage answers the user's question better than competing passages, and either quotes you (with a footnote link) or doesn't. There is no "position 3" — you're either cited or invisible.

Each engine runs a Retrieval-Augmented Generation (RAG) pipeline with two gates your page must clear:

1. Retrieval selection — can the engine find and fetch your page? This depends on crawler access (robots.txt), indexability, and topical relevance. Perplexity evaluates roughly 10 candidate pages per query.
2. Answer absorption — once retrieved, does your content actually shape the generated answer? Perplexity cites 3–5 of those 10 candidates. Pages that win are longer, more modular, and contain extractable evidence: definitions, numbers, comparisons, procedural steps.

Fail gate 1 and your content quality is irrelevant. Pass gate 1 but fail gate 2 and you were retrieved but not cited — the worst outcome because you can't diagnose it from access logs alone.

Gate 1: Crawler access (the prerequisite)

Each AI engine uses different crawlers. Block the wrong one and you're invisible to that engine regardless of content quality:

• ChatGPT: OAI-SearchBot (search/citations) + GPTBot (training) + ChatGPT-User (user-triggered fetches). You need OAI-SearchBot allowed for search citations.
• Perplexity: PerplexityBot — the most citation-heavy engine. Blocking it removes you from all Perplexity answers.
• Claude: ClaudeBot — used for training and retrieval for Claude's web-connected features.
• Google AI Overviews: Uses Googlebot (already allowed on most sites). No separate opt-in needed.

Copy-paste this into your robots.txt to allow all major AI citation engines:

User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

Our engine checks this automatically — run your URL through our AI crawler checker to see which bots can currently access your site.

Gate 2: Content structure that gets absorbed

Being retrievable isn't enough. Your content must be structured so the LLM can extract a clean, citable passage. We can be specific about which part of that most sites get wrong, because we measure it on every audit.

What 82 audited sites actually fail on

Every site run through our free checker is scored against the six checks that decide whether an answer engine can quote it. Across the 82 third-party sites audited to date (our own domains excluded, newest report per site, measured October 2026), the failures are not evenly spread — and they are not where most GEO advice points:

• 52 of 82 — few headings are phrased as questions a reader would ask
• 36 of 82 — no FAQPage or HowTo schema
• 30 of 82 — no llms.txt
• 20 of 82 — no Organization or WebSite schema
• 8 of 82 — no JSON-LD at all
• 6 of 82 — robots.txt blocks at least one AI crawler

The two at the top are the whole argument of this page. 52 of 82 sites write headings as labels rather than as the questions a reader would type, and 36 of 82 ship no FAQPage or HowTo schema. Those are precisely the two things that decide whether a model can find a passage and know what question it answers — the retrieval step and the attribution step. They are also both free: neither needs a plugin, a rewrite or a budget.

Meanwhile the thing GEO discourse argues about hardest — crawler access — is a non-problem in this sample: only 6 of 82 sites fail it. Almost nobody is blocked. Almost everybody is unquotable. If you are deciding where to spend an afternoon, that ratio is the answer.

Method: counts come from 101 completed reports collapsed to 82 unique hosts by ops/report-stats.mjs, snapshotted by ops/build-readiness-snapshot.mjs and imported into this page rather than retyped. The sample self-selects toward small and AI-adjacent sites, so treat every row as "of the sites we saw", not "of the web". Last measured October 2026.

Answer-first formatting

Put the direct answer in the first 40–60 words of the page (or section). AI retrieval systems weight opening content heavily — the first 200 words determine whether your page is relevant to the query. This also targets featured snippets and Google AI Overviews.

Bad: "In today's digital landscape, it's important to understand that canonical tags play a role in…" (50 words, no answer)
Good: "A canonical tag tells search engines which URL is the primary version of a page. Add it as <link rel="canonical" href="..."> in your <head>. Without it, duplicate pages split your ranking signals." (40 words, complete answer)

Question-phrased headings

AI engines map user questions to page sections by matching heading text. Headings phrased as questions ("How do I add a canonical tag in Shopify?") are retrieved more reliably than label headings ("Canonical tags").

Our engine flags this: the ai-question-heading-ratio check passes when at least 20% of your headings are question-like. Pages below 20% are consistently harder for retrieval systems to match to queries.

Chunk-friendly sections (under 400 words each)

RAG systems split pages into chunks for embedding and retrieval. Long, unbroken sections get split at arbitrary points, diluting the relevance signal. Short, heading-led sections (under 400 words each) stay intact as single chunks — each one can independently match a query.

Our engine measures this directly: the ai-content-chunkability check calculates average words between headings and flags pages over 400. If your average is 800+, you're making the retrieval system's job harder.

Extractable evidence types

Pages that get cited contain evidence the LLM can quote directly. Research from Princeton, Georgia Tech, and IIT Delhi found these patterns increase citation rates significantly:

• Statistics with attribution — "+40% citation rate" when passages include specific numbers with sources.
• Definitions — "X is Y" sentence patterns are 2.1× more likely to be cited than descriptive paragraphs.
• Step-by-step procedures — numbered steps are extracted as complete units.
• Comparison tables — structured data that answers "A vs B" queries directly.

Schema markup for AI citations

Structured data helps both Google AI Overviews and retrieval systems understand your content type. The three schemas that matter most for citations:

• FAQPage — maps questions to answers explicitly. Our engine's ai-faq-howto-schema-present check verifies this exists. AEO research weights FAQ schema quality at ~20% of citation likelihood.
• HowTo — marks procedural content as step-by-step instructions. Pairs with answer-first formatting for "how to" queries.
• Article — with datePublished and dateModified. Freshness is a top-5 signal for Perplexity; a visible date lets the retrieval system prefer your page over stale competitors.

Example FAQPage schema (copy-paste, fill in your content):

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "Your question here?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Direct answer in 1-2 sentences."
      }
    }
  ]
}
</script>

The llms.txt signal

llms.txt is a plain-text file at your domain root that tells AI models what your site is about, what content is most important, and how to use it. It's the AI equivalent of a sitemap — not required, but increasingly used by models during retrieval to prioritize which pages to consider.

Our engine checks for its presence (ai-llms-txt-exists). Generate yours with our free llms.txt generator — it takes under a minute.

Platform-specific timing

How long until changes appear in citations varies by engine:

• Perplexity: 2–7 days for structural changes. Fastest feedback loop — test here first.
• ChatGPT search: 7–21 days. OAI-SearchBot re-crawls on its own schedule; robots.txt changes take ~24 hours to propagate.
• Claude: 14–45 days. Longer cycle; focus on getting the fundamentals right rather than iterating.
• Google AI Overviews: Tied to Googlebot's crawl frequency. If you're already indexed, changes appear within 1–14 days.

Reputational signals (brand mentions, review-site presence, directory listings) take 30–90 days to propagate across all engines.

What our engine checks (your AI citation scorecard)

We built five specific checks into our engine that map directly to AI citation readiness. Run any URL through our free checker and you'll get a pass/fail on each:

1. llms.txt exists (ai-llms-txt-exists) — is there a machine-readable site summary for AI models?
2. AI crawlers allowed (ai-robots-allows-ai-bots) — are GPTBot, ClaudeBot, and PerplexityBot permitted in your robots.txt?
3. Question-heading ratio ≥ 20% (ai-question-heading-ratio) — are enough headings phrased as questions that retrieval systems can match?
4. FAQ/HowTo schema present (ai-faq-howto-schema-present) — does at least one page carry citation-friendly structured data?
5. Content chunkability < 400 words (ai-content-chunkability) — are sections short enough that RAG systems can extract them as clean, complete passages?

These five checks are specific, measurable, and directly tied to the mechanics of how AI retrieval actually works — not generic "write good content" advice.

The complete GEO checklist

Run through this for every page you want AI engines to cite:

1. robots.txt allows OAI-SearchBot, GPTBot, ClaudeBot, PerplexityBot
2. llms.txt exists at domain root with current site summary
3. First 40–60 words directly answer the target query (no preamble)
4. ≥ 20% of H2/H3 headings phrased as questions
5. No section exceeds 400 words between headings
6. FAQPage schema on pages with Q&A content; HowTo schema on procedural pages
7. Article schema with datePublished and dateModified (freshness signal)
8. At least one "evidence block" per section: a statistic, definition, comparison, or numbered procedure
9. Page loads in under 3 seconds (slow pages get dropped from retrieval candidates)
10. Content is substantive (1,500+ words for competitive queries — thin pages don't pass answer absorption)

FAQ

How long does it take to get cited by ChatGPT?

Structural fixes (schema, opening paragraph rewrites) typically appear in Perplexity citations within 2–7 days, ChatGPT within 7–21 days, and Claude within 14–45 days. Reputational signals like brand mentions and review-site presence take 30–90 days to propagate.

Does blocking GPTBot prevent ChatGPT from citing my site?

Blocking GPTBot prevents training, but ChatGPT search uses OAI-SearchBot. You need to allow OAI-SearchBot in robots.txt for your pages to appear in ChatGPT search citations. Blocking both means ChatGPT cannot cite you at all.

What content structure gets cited most by AI?

AI engines favor content with a direct answer in the first 40–60 words, question-phrased headings, sections under 400 words each, concrete data (numbers, dates, thresholds), and FAQPage or HowTo structured data.

Is GEO different from traditional SEO?

GEO (Generative Engine Optimization) targets AI answer engines instead of search result rankings. The overlap is large — structured content, authority, freshness — but GEO adds specific requirements: allowing AI crawlers, answer-first formatting, and chunk-friendly structure that retrieval systems can extract cleanly.