Explainer

Where does ChatGPT get its information when it searches the web

Where does ChatGPT get its information: from search providers such as Microsoft and OpenAI's own index, and it cites only some of the pages it reads.

Where does ChatGPT get its information when it answers with links: from outside search providers, Microsoft among them, and from an index of web pages that OpenAI crawls itself. OpenAI’s help center says ChatGPT search rewrites a question into one or more search queries, reviews the results and cites some of the pages.

TL;DR

  • ChatGPT search sends rewritten queries to outside search providers, Microsoft among them, and also uses an index of web pages that OpenAI crawls itself.
  • ChatGPT search cites fewer pages than it reads, and a 2026 AirOps study found ChatGPT left 85% of 548,534 retrieved pages uncited.
  • Website owners who want citations from ChatGPT search should allow the OAI-SearchBot crawler in robots.txt and in their host or CDN settings.

ChatGPT can search the web on its own when a question would benefit from current information, and a user can also choose Search by hand from the tools menu, according to OpenAI’s help center as of 29 October 2026. The split goes back to the launch of ChatGPT search on 31 October 2024, when OpenAI wrote that ChatGPT would choose to search based on what the user asks.

ChatGPT search typically rewrites a question into one or more targeted queries for its search partners, and after reviewing the first results it may send more specific ones, the help center says. In OpenAI’s example, a question about cancer drugs that target CCR8 becomes a search on CCR8 immunotherapy drug development, then a narrower one on CHS-114. The rewrite can also use the user’s approximate location and, when memory is on, saved memories.

Query fan-out is the practice of turning one question into several searches. In an AirOps study published on 12 March 2026, ChatGPT generated two or more fan-out queries on 89.6% of searches across 15,000 prompts.

ChatGPT, Bing and OpenAI’s own index

ChatGPT search draws on outside search providers, Microsoft among them, and on an index of web pages that OpenAI crawls and caches itself. At the launch on 31 October 2024, OpenAI wrote that ChatGPT search used third-party search providers plus content supplied directly by its partners. As of 29 October 2026, OpenAI’s help center names Microsoft, which runs Bing, and Shopify among the search providers that may process ChatGPT’s queries.

On whether ChatGPT uses Bing or Google, the citation data points to Bing. A Seer Interactive study published on 6 February 2025 found that more than 87% of citations from SearchGPT, the prototype that became ChatGPT search, matched Bing’s top 20 organic results for the same question, against 56% for Google. Seer called its sample of about 100 questions directional.

Christina Blake, Director of Product, and Alisa Scharf, Chief AI Officer, at Seer Interactive wrote that “for longer or more complex questions, ChatGPT likely simplifies the query.”

OpenAI’s help article on offline web search says some business workspaces can run ChatGPT web search only on OpenAI’s indexed and cached copy of the web.

ChatGPT sources and citations

ChatGPT search reads more pages than it cites, and the cited pages appear as links in the answer and in a Sources panel. OpenAI’s help center says a user can select a citation to open its source, or select Sources to see the cited sources and other relevant links. OpenAI’s documentation for its web search tool, which developers build into their own apps, lists every URL the model consulted separately from the citations, and sources often outnumber citations.

ChatGPT left 85% of the 548,534 pages it retrieved uncited in AirOps’ study of 15,000 prompts, published on 12 March 2026. OpenAI’s help center says ChatGPT ranks search results using multiple factors meant to help users find relevant, reliable information, and that placement is not guaranteed. The help center does not list those factors. Retrieval is the first step toward getting a brand cited in ChatGPT answers, and the ranking decides which retrieved pages earn the link.

What publishers can control

OpenAI’s ChatGPT search publisher guidance centers on one crawler, OAI-SearchBot, and as of 29 October 2026 it asks sites that want to appear to let that crawler in. OpenAI’s help center tells site owners to allow OAI-SearchBot and to confirm that their host or content delivery network (CDN), the service that serves the site’s files, accepts traffic from OpenAI’s published IP addresses for the crawler.

OpenAI’s crawler documentation says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links. It adds that a robots.txt change can take about 24 hours to reach search results.

OpenAI’s help article on offline web search lists why pages can be missing from OpenAI’s index: robots.txt blocks, CDN or bot-blocking rules, login walls, content that depends on scripts, and pages that are new or rarely accessed. These access settings are the base layer of generative engine optimization, the practice of shaping pages so AI assistants retrieve and cite them.

Access is the first condition, and new pages still wait days or weeks for a first citation, as how long GEO takes to show up in ChatGPT answers measures.

Checking whether ChatGPT cites your site

A brand can see whether ChatGPT search cites its site by asking ChatGPT the questions its buyers ask, with search on, and opening the Sources panel under each answer. OpenAI’s help center says that panel lists the cited sources and other relevant links, and it warns that search results and citations can be incomplete, outdated or incorrect. A cited page is worth opening to see whether the answer reports it correctly.

Server logs show the crawler side. OpenAI’s crawler documentation gives the user agent of each crawler, including OAI-SearchBot, which crawls pages for ChatGPT’s search features, and ChatGPT-User, which may visit a page when someone asks ChatGPT a question. OpenAI publishes the IP addresses of both, so a site team can check that a visit came from OpenAI. The Sources panel and server logs are two methods for checking whether an AI assistant has cited your site.

Asking the same buyer questions each month and counting how often the brand is named turns these checks into a measure of AI visibility.

What it means

Website owners and SEO leads who want ChatGPT search to cite their pages should fix crawler access first and then check the answers, in this order.

  1. Allow OAI-SearchBot in robots.txt, and ask your host or CDN to accept OpenAI’s published IP addresses for the crawler.
  2. Wait about 24 hours after a robots.txt change, the time OpenAI gives for its systems to adjust.
  3. Put your buyers’ questions to ChatGPT with search on, open Sources and list every domain cited.
  4. Search the same questions on Bing, whose top results matched most ChatGPT search citations in Seer Interactive’s 2025 study.
  5. Repeat the questions each month and note whether your pages move into the cited list.

FAQ

No. OpenAI’s crawler documentation says each setting is independent, so a site can allow OAI-SearchBot to appear in search results while disallowing GPTBot, the crawler that collects content that may be used to train OpenAI’s models. ChatGPT-User, the agent behind visits that users trigger, is not used to decide whether content appears in search either.

Anyone who uses ChatGPT can search the web, including people who are not signed in. OpenAI’s help center lists web search on the Free, Go, Plus, Pro, Business, Enterprise and Edu plans as of 29 October 2026, within each plan’s usage limits. OpenAI opened ChatGPT search to everyone without signup on 5 February 2025, in regions where ChatGPT is available.

Is there data on how often a brand’s own website is cited?

Yes: a 2026 ChatGPT, Perplexity and Google AI Overviews citation study checked each brand’s own website against its category’s top citations and gave the percentage missing from the sources behind AI answers. 34.8% of the businesses in Agenzy’s 2026 source study did not have their own website among the 30 sources AI engines cited most for their own category (n=138 business projects). The ranks by engine are in how often LLM citations include a brand’s own website.

Sources

  1. help.openai.com/en/…/9237897-searching-the-web-with-chatgpt
  2. openai.com/index/introducing-chatgpt-search
  3. developers.openai.com/api/…/bots
  4. help.openai.com/en/…/20001203-offline-web-search-for-chatgpt-work…
  5. developers.openai.com/api/…/tools-web-search
  6. seerinteractive.com/insights/87-percent-of-searchgpt-citations-match-bing…
  7. airops.com/report/influence-of-retrieval-fanout-and-google-ser…
  8. agenzy.lt/blog/which-sources-ai-engines-cite-2026

The week in AI search, in one email

One email a week with what changed in ChatGPT, Google AI Mode, Perplexity and Gemini, and the week's data piece.

Up next