Guide

How to get cited by ChatGPT: what actually decides it

The mechanism behind a ChatGPT citation, the factors you can actually control, what does not work, and how to check your own site.

By Sunny Patel · Updated 2026-09-20

A ChatGPT citation happens when ChatGPT search names or links your page as the source for part of an answer. It runs on a mechanism most sites never check directly: OpenAI's own search crawler, plus a retrieval step that also draws on third-party search providers and other sources already covering the same topic. A page that mechanism cannot crawl, or cannot find, never enters the running regardless of how well written it is. This guide covers that mechanism first, then the factors you can actually influence, what does not work, and how to check your own site.

ChatGPT search draws on more than its own crawl. OpenAI runs a dedicated crawler for this purpose, OAI-SearchBot, which is a different agent from GPTBot: GPTBot gathers training data for OpenAI's models, while OAI-SearchBot builds the index ChatGPT search actually cites answers from. Blocking one has no effect on the other. OpenAI's own documentation also states that ChatGPT search draws on third-party search providers alongside its own crawl, without naming which ones. A page blocked from OAI-SearchBot, or absent from whichever third-party index feeds a given query, is missing from one or more of the pools ChatGPT search retrieves from before it ever reaches the step where content quality gets judged.

Why a blocked or unindexed page cannot be cited, whatever it says

Citation is the last of three separate checkpoints, and a page has to clear all three in order. It has to be crawlable: robots.txt allows the relevant bot and a working link path leads into the page. It has to be indexed: the crawler has actually processed the page and added it wherever that system stores what it knows. Only then does citation become possible: a real query arrives, and the retrieval step judges this page a good enough match to name as a source. A page that is well written but blocked, or well written but never crawled because nothing links to it, never reaches the third checkpoint at all.

The orphan pages guide covers the specific failure where a page exists and returns a normal response but has no internal link path leading a crawler to it in the first place, which is one of the most common reasons a page never gets to the indexed stage.

A site that renders its content entirely with client-side JavaScript runs into the same wall earlier. OAI-SearchBot, like most AI and search crawlers, reads raw HTML and does not execute a full browser render. Content that only appears after JavaScript runs is functionally invisible to it, even though a human visitor with a browser sees it fine.

What you can actually control

None of the following guarantees a citation. Each one removes a specific reason a citation cannot happen, or improves the odds a retrieval step picks your page over a competitor's for a given question.

Crawlable raw HTML with real content

The content that decides a citation is the raw HTML bytes OAI-SearchBot downloads, not the version a visitor sees after scripts run. A page needs its actual claims, numbers and explanations present in that raw response. Thin content padded around a single fact reads the same to a retrieval system as it does to a person skimming it: not worth citing over a page that states more clearly.

Clear entity and topic ownership

A retrieval step favours a page that is unambiguous about what it covers and who it belongs to. A page that clearly states its subject, in a direct heading and opening paragraph, before any preamble, gives a retrieval system less work to do in deciding whether it matches a query. A page that hedges, buries its actual claim several paragraphs in, or covers several unrelated topics at once, is harder for that same system to match confidently against a specific question.

Being present in the sources ChatGPT already pulls

OpenAI documents that ChatGPT search draws on third-party search providers alongside its own crawl, without naming which ones. Being indexable by a major search engine such as Bing costs nothing to check and may help, but treat any specific benefit to ChatGPT citations as indirect and unconfirmed, not a guaranteed input. What is more clearly useful is the conventional off-page signal underneath it: a page mentioned and linked by other sites covering the same topic gives a retrieval step more places to find the same claim corroborated, regardless of which index happens to carry it.

Answer-first pages

A page that states its answer plainly near the top, before qualifications and background, matches how a retrieval step extracts a citable claim. A page structured as a long narrative build-up towards a conclusion buried at the end makes that extraction harder, even when the underlying information is the same. Definition-first, answer-first structure is the same principle this site's own guides follow throughout.

Structured, checkable facts

A specific figure, a named source, or a dated claim gives a retrieval system something concrete to cite, not just a general impression. Vague, unfalsifiable statements are harder to cite confidently because there is nothing precise to attribute. This does not mean inventing statistics to look precise. It means stating the real facts you do have as specifically and plainly as the underlying evidence actually supports.

What does not work

An llms.txt file alone does not secure a citation. It is a community proposal, not a confirmed input to OpenAI's production crawlers, and no major AI lab has publicly confirmed its search crawlers read or act on it. It costs little to add and does no harm, but it never substitutes for a page OAI-SearchBot can actually crawl and a retrieval step can actually match to a query.

Keyword stuffing does not work either, and tends to actively hurt. A phrase repeated without adding information reads as padding, not a clear claim, to a retrieval system and to a human reader checking the citation afterwards. The same applies to publishing thin pages purely to cover more keyword variants: a retrieval step favours one page that answers a question directly over several shallow pages that each half-answer it. The llms.txt guide covers the format and its current evidence in full.

How to check your own site

1. Check robots.txt for a rule that blocks OAI-SearchBot specifically. A rule aimed only at GPTBot has no effect on search citations, since the two are separate agents with separate rules.
2. Run sitemap.digital's free AI crawler access checker to see, in one pass, whether OAI-SearchBot and the other major AI crawlers can currently reach the page.
3. Run a full scan and check the page's AI-readiness parts, covered in the AI crawlability score guide, for the structural issues, thin content, missing schema, deep click paths, that make a page harder to crawl and harder to match to a query even once access is confirmed.
4. For the full picture across a whole site, not one page at a time, run the site audit, which brings crawlability, structure and AI readiness together in one report.

None of this promises a citation on any specific query. What it does is remove the mechanical reasons a citation cannot happen at all, and put your page in a fair position to compete on the content itself, instead of losing before that competition even starts.

Frequently asked questions