There is no public “ChatGPT ranking algorithm.” OpenAI has never published a complete description of how ChatGPT search picks the pages it cites, so anyone who tells you the exact formula is guessing. What we can do is separate what is documented from what practitioners have measured, and use both to decide what to change on your site.
What is documented: three crawlers
OpenAI documents three separate user agents, and you can control each one in your robots.txt file:
- GPTBot collects content that may be used to train future models.
- OAI-SearchBot builds and refreshes the index behind ChatGPT search results. Blocking it can keep your pages out of those results.
- ChatGPT-User fetches a specific page in real time when a user’s conversation needs it.
Many sites block all three with one blanket rule and then wonder why they never appear. If you want to be cited, allow OAI-SearchBot at minimum.
How a question probably becomes a citation
Practitioner analyses describe a pipeline like this. It is our reading of observed behavior:
- Routing. The system decides whether the question needs live web data or can be answered from training.
- Query fan-out. A conversational prompt is split into several shorter searches, often three to eight, covering different angles.
- Retrieval. Tens to hundreds of candidate documents come back from a search index. Bing supplied the original index, but it is no longer the only source.
- Reranking. A smaller model scores candidates for relevance and reliability and narrows them to a handful.
- Synthesis. The answer is written, with citations attached to supporting passages.
The practical lesson is that a page can be lost at any stage for a different reason: not crawled, not retrieved, not selected or not quoted. Bing rank correlates only loosely with being cited.
What tends to get cited
| Finding | Detail | Source |
|---|---|---|
| Only some retrieved pages are cited | ChatGPT cites roughly 15% of the pages it retrieves | Zyppy, via Kime |
| The top of the page matters | 44.2% of citations come from the first 30% of a page | Zyppy, via Kime |
| Third-party mentions carry weight | Earned media outperforms brand-owned content for citation rates by a wide margin in one study | AuthorityTech, via Kime |
| Structure helps | Pages with FAQ schema received about three times more citations than unstructured equivalents | Authoritas, via Kime |
| Reference sources dominate | Wikipedia accounted for about 12% of 83,670 analyzed citations; LinkedIn about 4% | Analyze AI, via Kime |
Where on the page ChatGPT cites from
Share of citations by position in the source page.
- First 30% of the page
- 44.2%
- Middle of the page
- 31.1%
- Final 30% of the page
- 24.7%
Source: Zyppy (2025), as reported by Kime. Position is measured as a share of the page’s length.
What to do with this
- Check robots.txt. Make sure OAI-SearchBot is allowed on the pages you want cited.
- Answer one question per page or section, in the first two or three sentences.
- Include specifics such as numbers, dates and names that can be lifted without paraphrasing.
- Show dates. Visible publication and update dates help on fast-moving topics.
- Earn corroboration. Claims that also appear on other credible domains are easier for a model to trust.
- Stay realistic. Reddit’s share of citations reportedly swung from about 60% to about 10% in a few weeks in 2025. Sources shift, so measure regularly.
For the broader playbook, read our guide to getting cited by ChatGPT, Perplexity and AI Overviews, or learn about generative engine optimization. If you would like us to check your robots.txt and your top pages, get in touch.
Sources: Kime (summarizing Zyppy, Authoritas, AuthorityTech and Analyze AI) and Lemniscate Growth, reviewed in September 2026. Third-party studies use different methods; treat them as directional.
Related reading
Share this article