How to get cited by AI search

When someone asks ChatGPT, Perplexity or Google AI Overviews a question, the model does not invent an answer from nothing. It retrieves a small set of sources, grounds its answer in them and names the ones it leans on. Getting cited means being one of those named, linked sources.

This is different from classic SEO. You are not fighting for a blue link on page one. You are trying to be the sentence an AI quotes and the domain it attributes. The unit of visibility is the fact, not the page: a specific, quotable, source-backed claim that a model can lift with confidence.

Every generative engine runs a version of the same loop: retrieve candidate documents, rank them, ground the answer and attribute. Five properties push you up that ranking.

• Retrievability. Your content has to be crawlable and readable. AI crawlers such as GPTBot, ClaudeBot, PerplexityBot and Google-Extended must be allowed, and the fact has to sit in clean HTML, not locked inside an image or a script.
• Quotable specificity. Named person, concrete number, clear claim. Models prefer a source they can quote verbatim over vague prose they would have to paraphrase.
• Structured signals. Schema.org markup, a clear entity, a machine-readable export. The easier a fact is to parse, the more confidently it is cited.
• Authority and corroboration. A claim that is attributed to a named source and echoed elsewhere is safer to cite than an unsourced assertion.
• Freshness. For anything time-sensitive, recency wins. A dated, clearly published fact beats an undated one.

ChatGPT surfaces sources through its search feature and the OpenAI crawlers (GPTBot for training and indexing, OAI-SearchBot and ChatGPT-User for live answers). To be cited, allow those agents in your robots.txt, publish the fact in clean HTML and make the claim self-contained so it reads correctly out of context.

Structured data and a named attribution help ChatGPT pick you as the source rather than a competitor covering the same story.

Perplexity is the most citation-first of the major engines: it shows numbered sources next to almost every answer. It rewards pages that answer a specific question directly, with a quotable line near the top and a clear publication date. Allow PerplexityBot and Perplexity-User, keep the answer above the fold and attribute the claim to a named source.

Google AI Overviews draw on the same index as classic search, then synthesise an answer and cite a few pages. Ranking still matters, but the pages that get pulled into the overview tend to answer the question plainly, use clear headings and carry strong structured data. Allow Google-Extended, mark up your facts with schema and write the answer as a direct, standalone statement.

Claude fetches live sources through ClaudeBot and Claude-User and increasingly through its own search. It favours clean, well-attributed, source-backed content and machine-readable formats. Allow the Anthropic agents, keep your facts in plain HTML and pair each claim with its original source so Claude can verify before it quotes.

Across every engine the same short list moves the needle.

1. Allow the AI crawlers (GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended and friends) in robots.txt.
2. Put each fact in clean, crawlable HTML with a named speaker and a concrete number.
3. Add schema.org structured data and a machine-readable export such as JSON.
4. Link every claim to its original source so it can be verified.
5. Keep it fresh and clearly dated.
6. Get the fact indexed somewhere built for AI retrieval so engines encounter it in a citation-ready shape.