OpenAI documents the broad process: ChatGPT can turn your prompt into search queries, retrieve results, rank them using multiple factors and write an answer with links. It doesn't publish those factors or their weights. So any exact list of what ChatGPT prefers is an inference, however confidently it's presented.
What does OpenAI document?
OpenAI documents the steps and the access rules, not the selection formula. These points come from its search help page, its crawler overview and its publisher FAQ.
- Queries. ChatGPT can turn a request into one or more search queries and use the results to write an answer.
- Partners. It sometimes works with other search providers, and usually sends them rewritten, targeted queries.
- Personal context. With memory on, saved memories may shape those queries. A general location estimated from the IP address may be shared with a provider.
- Ranking. Results are ranked using multiple factors meant to find relevant, reliable information. Placement is not guaranteed.
- Showing sources. Answers can carry inline links. Sidebar sources can appear without being directly cited, and with third-party providers their order is influenced by that provider's ranking.
- Eligibility. A site opted out of OAI-SearchBot won't be shown in search answers, though it may appear as a navigational link.
OpenAI's Enterprise and Edu page names Bing as a search engine that can receive disassociated queries in those workspaces. Don't assume that page describes every plan or will stay the same.
What doesn't OpenAI publish?
It doesn't publish the list of factors, how they're weighted, which share of answers use web search, or why one specific source was picked over another. We also found no statement in the pages above that domain age, word count, publishing frequency or markup helps. If someone gives you a weighted list, ask them where it comes from.
What can we reasonably infer?
| Inference | Based on | Could be wrong because |
|---|---|---|
| Doing well in regular search helps | Partner providers and their ranking influence on the sidebar | OpenAI also runs its own crawler, and the mix isn't public |
| Pages must match the rewritten query | Queries are rewritten into targeted terms | We rarely see the actual queries |
| Clear statements are more likely to be used | An answer needs something it can state and cite | Not confirmed as a factor by OpenAI |
| Independent mentions matter | Answers cite third-party pages | Being mentioned isn't the same as being trusted |
Third-party studies of which domains get cited can be useful, but treat them with caution. They show correlation in a sample of questions at a moment in time, and it's worth checking who published them and what that company sells. They don't reveal OpenAI's criteria.
How can you learn what applies in your market?
Your own evidence is better than a generic list. Run a fixed set of buyer questions, open the sources behind each answer and note the type of site each one is. In one market that may be trade media. In another it may be comparison sites or forums. Then ask one question per source: do we belong here, and is what it says about us correct?
What does this mean for your site?
- Confirm that OAI-SearchBot can reach your site, including through your host or CDN
- Treat regular search visibility as the base, not as something separate
- Write the facts you want repeated in plain text on the page
- Correct what third-party sources get wrong
- Treat every claim about ChatGPT's preferences as a hypothesis until you've tested it
For the practical side, read how to approach ranking in ChatGPT. Our free scan is one way to check how readable your site is.
