本站有中文版。
切换到中文The answer engine most buyers outside China reach for first.
We ask it your buyers' questions with web search on, so the answer cites live pages, and record which pages it cites and which brands it names. We ask through both the API and ChatGPT Search on the web, because the two can answer differently.
With web search on, ChatGPT fetches pages with its own crawler, OAI-SearchBot, and cites only what it could load. A page blocked in robots.txt or slow enough to time out is not a candidate. The AI crawler check on this site tells you which crawlers your robots.txt lets in.
The answer is assembled from passages. A page that makes its claim in one clear sentence gives the engine a passage to use. One that buries the point after three paragraphs of preamble gives it nothing to quote.
With search off, ChatGPT answers from what it read in training, which updates slowly. A name that appears consistently in comparisons, reviews and directories comes up. One that lives only on its own site does not.
Both, and they behave differently. With web search on it retrieves pages and cites them, so a new page can be cited the week it goes live. With search off it answers from training data. Source asks both ways and keeps the two sets of answers apart.
OpenAI sells ads inside ChatGPT, bought and labelled as ads. The recommendation inside the answer itself is not for sale.