Insights · Field note · Growth and AI visibility
How ChatGPT chooses which pages to cite
What is observable about retrieval and citation in ChatGPT, what is documented by OpenAI, and what that implies for a page that wants to be named.
ChatGPT cites pages when it searches, which it does when it judges the question needs current or specific information or when the user asks it to. Its search crawler, OAI-SearchBot, must be able to fetch the page. Among fetched candidates it favors passages that answer the question directly, from sources it can identify, and it shows little regard for Google rank: Semrush found almost 90 percent of ChatGPT-cited pages at Google position 21 or worse. The practical levers are crawl access, a direct answer, verifiable attribution and consistency.
In detail
What the evidence shows.
Retrieval is conditional
ChatGPT does not search for every prompt. It searches when the model judges that the answer depends on current or specific information, or when search is requested. Questions about providers, prices, regulations and recent events trigger it; general explanations often do not. A page can only be cited on prompts where retrieval happened, which is why the prompt set should include the specific, decision-stage questions that trigger it.
The crawler must be allowed
OpenAI documents separate agents: GPTBot for training data, OAI-SearchBot for search results, and ChatGPT-User for fetches the user initiates. A site that blocks OAI-SearchBot, deliberately or through a hosting default, cannot appear as a source in ChatGPT search answers regardless of quality. The server logs show whether the crawler arrives; nothing else does.
Rank is not the selector
Semrush’s analysis found almost 90 percent of ChatGPT-cited pages ranked 21st or worse on Google for the same query. The retrieval system is choosing the passage that best answers the question from a source it can attribute, not the page Google ranks highest. Pages built to rank, broad and keyword-led, are often passed over for pages built to answer.
What the cited pages have in common
In our citation tracking across client sites (first-hand: David Selakovic), cited pages share a declarative answer in the first screen, a named author the engine can verify, consistent entity facts, and currency. None of those is a trick; each is a property of a page written to be the source. See what makes a page citable.
Sources
Primary references.
- OpenAI: overview of OpenAI crawlers (GPTBot, OAI-SearchBot, ChatGPT-User)
- Semrush, 2025. AI Mode and AI Overviews study: about 69 million U.S. desktop sessions, May to July 2025; AI Overviews trigger rate measured November 2025; AI-search visitor conversion and ChatGPT citation position analysis.
By David Selakovic · Published 2026-10-16 · Revised 2026-10-16 · Reviewed every 180 days