Most ChatGPT answers come from training data
ChatGPT ran a web search on 34.5% of queries in February 2026, according to Semrush clickstream data covering October 2024 to February 2026. In late 2024 the share was 46%. Every other answer comes from what the model learned in training.
OpenAI says its models are trained on three sources: information publicly available on the internet, information from third-party partners, and information that users, human trainers and researchers provide. Public web content includes webpages, forums, blogs and posts.
“For publicly available internet content, we use only information that is freely and openly accessible on the internet.” OpenAI
GPTBot is the crawler that collects this training content. Blocking it keeps your future pages out of training and does not affect ChatGPT search. See GPTBot for the robots.txt rules.
ChatGPT search uses three sources of web results
When ChatGPT searches, it rewrites the question into one or more searches. OpenAI’s ChatGPT search help page says it “sometimes partners with other search providers” and sends them those searches. The page lists the Microsoft privacy statement for those providers. OpenAI’s Enterprise and Edu page says ChatGPT “may share disassociated search queries with the Bing search engine.” OpenAI does not name Google anywhere in its documentation.
OpenAI also crawls for its own index with OAI-SearchBot. OpenAI says sites that block OAI-SearchBot will not be shown in ChatGPT search answers, apart from navigational links.
Four independent studies measured which source ChatGPT’s citations come from:
| Study | Sample | Finding |
|---|---|---|
| Seer Interactive, February 2025 | About 100 questions, 500+ citations | 87% of ChatGPT search citations matched Bing’s top 20 results, against 56% for Google’s. |
| Backlinko test, 2025 | One page that only Google could crawl | ChatGPT Plus with browsing answered from the page. The free version did not. |
| RESONEO, August 2026 | 1,200 answers, 88,000 results | Free Think mode took 74.7% of results from OpenAI’s own index and 3.1% from Google. Paid Thinking took 75.3% from scraped Google results and 24.7% from OpenAI’s index. |
| Ahrefs | 118,931 fan-out searches | Only 6.82% of ChatGPT search results were in Google’s top 10, and 83.39% did not appear in Google’s results at all. |
The studies disagree because they measured different modes at different times. ChatGPT’s sources have changed more than once since 2024, and none of these numbers will hold for long.
Does ChatGPT use Bing or Google? Plan for both
ChatGPT uses Bing, Google results and OpenAI’s own index, in proportions that change by mode and over time. Optimising for one of them is a bet. A page that is crawlable by OAI-SearchBot and indexed in both Google and Bing is in every pool ChatGPT draws from.
The same applies to pages on other sites that mention you. A review or roundup that names your product can only show up in a ChatGPT search if one of those sources has indexed it.
How to get your pages into ChatGPT’s sources
- Allow OAI-SearchBot in robots.txt and at your firewall. Check it with the AI crawler checker. OpenAI says robots.txt changes take about 24 hours to reach ChatGPT search.
- Get the page indexed in Google and Bing. Index My Site sends a URL to Google for 1 credit, or to Google and Bing for 2.
- Get the pages that mention you indexed as well. See why unlinked brand mentions matter for AI search.
- Check whether ChatGPT cites you for a question with the AI visibility checker, then read how to get cited by ChatGPT.
Start indexing to send any public URL to Google and Bing.