How to get cited by ChatGPT and Perplexity (and what counts as a citation)

Two years ago, ranking #1 on Google was the goal. Today there's a parallel game: being the URL ChatGPT, Claude, and Perplexity link to when a user asks a question in your space. The mechanics are different from search SEO, and the people who figure it out early are eating a growing referral channel for free.

How LLM citations actually work

When a user asks a question, the LLM either pulls from its training data (no citation) or runs a real-time retrieval (search) and grounds the answer in fetched documents. Citations only appear on the second path. That means citation-driven traffic depends on (a) your URL being in the retrieval index, and (b) the answer model deciding your URL is the right thing to cite.

What the retrieval indexes use

• Perplexity has its own crawl and a Bing partnership underneath.
• ChatGPT Search uses Microsoft Bing as the primary retrieval index, plus OpenAI's own OAI-SearchBot crawl.
• Claude uses Anthropic's own crawl plus Brave Search (for Claude-Web fetches).
• Google AI Overviews use the regular Google index — same one that ranks Search.

What gets cited (the patterns)

Across LLMs, a few consistent patterns emerge for what gets picked when there are multiple candidate URLs to cite.

1. Specificity over comprehensiveness. A page that directly answers the exact question wins over a longer page that touches on it.
2. Recency. LLMs prefer URLs dated within the last 12-18 months for time-sensitive topics.
3. Structured data. Pages with FAQ schema, HowTo schema, or clear Q&A formatting are easier to extract clean citations from.
4. Clean H1 + first paragraph. The first 200 words after the H1 are what the LLM sees first when it ranks candidates.
5. Source authority. Trusted domains (Wikipedia, official docs, .gov, established publishers) get disproportionate citations even when smaller sites have better content.
6. Direct quotability. If your page contains a sentence that directly answers the question in clean prose, you'll get cited verbatim.

Write for citation extraction

If you're optimizing for LLM citations, the page structure that wins looks different from a typical SEO blog post. Imagine the LLM is trying to lift one paragraph from your page and quote it in an answer — write that paragraph.

• Open with a direct, self-contained answer to the page's core question. No preamble.
• One key claim per paragraph. Each paragraph should be liftable in isolation.
• Use specific numbers and dates. "Google's Indexing API allows 200 publish requests per project per day" beats "a generous daily quota."
• Add an FAQ section. LLMs disproportionately cite FAQ blocks because the Q→A structure matches their retrieval patterns.
• Cite your own sources. LLMs that pull your page see your citations as quality signals.

Add structured data the LLMs read

Schema.org JSON-LD remains the cheapest way to tell crawlers (search and AI) what your page is about. Helpful types for citation extraction:

• FAQPage — pages with Q&A blocks.
• HowTo — step-by-step instructions.
• Article + datePublished — recent timestamping wins.
• DefinedTerm — glossary entries, especially useful for technical concepts.
• Person / Organization with sameAs — author / publisher entity reconciliation.

Make sure the answer-time bots can reach you

All of the above is wasted if your robots.txt blocks ChatGPT-User, PerplexityBot, or Claude-Web. The single most common mistake in 2026: site owners block GPTBot for training, accidentally extend it to all OpenAI bots, and lose the ChatGPT citations they wanted to keep.

Submit to Bing too

Since ChatGPT and Perplexity both lean on Bing's index, IndexNow submissions to Bing are now meaningful for AI citation. IndexerNow's Bing tab pushes URLs to Bing Webmaster + Yandex via IndexNow with the same paste-box flow as Google.