Research
How ChatGPT search works: a GPT-4o model in front of Bing
ChatGPT search does not run its own web index. By OpenAI's own account it is a fine-tuned GPT-4o model that rewrites your question, sends the queries to third-party search providers including Bing, and writes up the results with citations. That single fact changes what getting cited actually requires. Every claim below links to OpenAI's or Microsoft's own words.
Key takeaways
- ChatGPT search is “a fine-tuned version of GPT-4o” that “leverages third-party search providers.”[1] No primary source documents OpenAI running its own general web index.
- Those providers include Bing and Shopify, named in OpenAI's help center;[2] Microsoft states plainly that “Bing powers… AI tools like ChatGPT.”[4]
- ChatGPT “rewrites your query into one or more targeted queries” and refines them iteratively before answering.[2] You are matched on the rewrite, not your literal words.
- OAI-SearchBot, not GPTBot, controls inclusion: sites opted out of OAI-SearchBot “will not be shown in ChatGPT search answers.”[3] Blocking GPTBot only opts you out of training.
- OpenAI discloses almost nothing about ranking, only “a number of factors” and “no way to guarantee top placement.”[2] So the levers that remain are being in Bing and being cleanly crawlable.
In this article
People optimize for “ChatGPT” as if it were a search engine. It is not one. It is a language model with a search tool bolted on, and the search tool calls someone else's engine. Once you see that, the advice that actually moves the needle stops being about the model and starts being about which index it reads and which crawler it obeys. Here is what OpenAI and Microsoft have put in writing.
What ChatGPT search actually is#
OpenAI says it plainly. In the launch announcement, ChatGPT search “leverages third-party search providers, as well as content provided directly by our partners,” and the search model is “a fine-tuned version of GPT-4o.”[1] A model, and a rented search. It grew out of the SearchGPT prototype announced in July 2024 and launched as ChatGPT search on October 31, 2024.[5][1]
The consequence of that architecture is the through-line of this whole piece. A model cannot cite a page a search provider did not return, and a search provider cannot return a page that is not in its index. So the work of getting into ChatGPT answers happens almost entirely at the search layer, which OpenAI does not own. This is the opposite of Perplexity, which built its own index, a difference we laid out in the index machines behind ChatGPT, Claude, Perplexity, and Google.
How your question becomes searches#
ChatGPT does not search for the words you typed. OpenAI's help center says the model “rewrites your query into one or more targeted queries that it sends those providers,” and that after reviewing the first results it “may send additional, more specific queries” to refine the answer.[2] One question can become several searches, tightened over a couple of rounds.
For your content that means two things at once. You are found through the rewritten query, so covering the vocabulary a buyer's question would be rewritten into matters as much as covering the substance, and because the model iterates toward specificity, a page that answers a narrow, specific version of the question directly can be pulled in on the second or third search even if it loses the broad one.
The Bing dependency#
The named providers are not a secret. OpenAI's help center points users to the privacy policies of “Bing (Microsoft)” and “Shopify” as the third-party search providers ChatGPT sends queries to.[2]Microsoft confirms it from its own side: “Bing powers a wide range of AI and search experiences, including Yahoo, DuckDuckGo, Microsoft Start, and AI tools like ChatGPT and Inflection.ai.”[4] No primary source shows OpenAI operating its own general web index. The retrieval is rented.
This is not a fringe arrangement, it is a market. Microsoft retired the public Bing Search APIs on August 11, 2025 and pushed customers to “Grounding with Bing Search as part of Azure AI Agents” instead, moving index-as-a-service up the stack toward AI agents.[6]The practical takeaway for you is unglamorous and load-bearing: if you are not in Bing's index, ChatGPT search has nothing of yours to cite, no matter how good the page is.
the two third-party search providers OpenAI names in its help center for ChatGPT search. Your presence in Bing's index is a precondition for being cited.
The crawler that controls inclusion#
OpenAI runs three crawlers with distinct jobs, and confusing them is the most common self-inflicted wound in AI visibility. Its documentation states that GPTBot “is used to crawl content that may be used in training our generative AI foundation models,” while “OAI-SearchBot is used to surface websites in search results in ChatGPT's search features.”[3]
The line that matters is unambiguous: “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers.”[3]ChatGPT-User, the agent that fetches a page when a user asks, is separately documented as “not used to determine whether content may appear in Search.”[3]So blocking GPTBot is a training decision with no effect on search visibility, and blocking OAI-SearchBot quietly deletes you from ChatGPT's answers. Sites that add a blanket AI-bot block to “opt out of AI,” or whose CDN does it for them, routinely remove themselves from ChatGPT search without meaning to. This is the first thing we check in an audit.
What OpenAI does not tell you#
Here honesty requires saying what is not knowable. Unlike Perplexity, OpenAI publishes no index scale, no retrieval architecture, and no ranking algorithm. Its help center offers only that “ranking in ChatGPT Search is based on a number of factors designed to help users find reliable, relevant information,” and that “there is no way to guarantee top placement.”[2] Anyone selling you a precise ChatGPT ranking formula is guessing.
What you can act on is the part that is documented: the index is Bing's, the gatekeeper crawler is OAI-SearchBot, and the match happens on a rewritten, increasingly specific query. Optimization is making sure each of those three known stages can find you, rather than chasing an undisclosed scoring function.
How to get cited in ChatGPT#
Everything above reduces to four moves, each tied to a documented mechanism.
1. Be in Bing's index
ChatGPT queries Bing.[2][4] If Bing has not indexed a page, ChatGPT search cannot cite it. Verify coverage in Bing Webmaster Tools and fix what Bing cannot crawl, before touching anything ChatGPT-specific.
2. Let OAI-SearchBot in
Allow OAI-SearchBot in robots.txt and at your CDN or firewall, and keep GPTBot as a separate, deliberate decision.[3] Opting out of training should never cost you search inclusion, but a careless bot rule makes it do exactly that.
3. Answer the rewritten query directly
Because the model searches a specific rewrite of the question,[2]structure pages so a section states the question and answers it plainly, and cover the category vocabulary that a buyer's question would be rewritten into. Retrievability beats cleverness.
4. Give the answer citable evidence
The founding GEO research found that adding concrete statistics, direct quotations, and citations to authoritative references measurably raised how often a source was surfaced in generated answers.[7] An engine whose product is a cited answer favors passages that carry specific, verifiable claims.
One caution before you pour everything into this engine: winning ChatGPT tells you little about the rest. Across 100,000 identical prompts, ChatGPT and Perplexity shared only 11 percent of their cited sources.[8] Different index, different result.
Measuring whether ChatGPT cites you is the easy part, and the part most tools stop at. The useful part is fixing why it does not: thin coverage in Bing, an OAI-SearchBot block, or a passage that does not answer the rewritten query. Noetio finds which of those is the cause, ships the fix, and re-measures to confirm the citation changed, not just re-scores it. How that differs from the monitoring platforms is laid out, with sources, on the compare page.
Not cited by ChatGPT? See exactly why, and what fixes it.
Sources
- ChatGPT search “leverages third-party search providers”; the search model is “a fine-tuned version of GPT-4o”: OpenAI, “Introducing ChatGPT search”, October 31, 2024.
- ChatGPT “rewrites your query… sends those providers” and refines iteratively; names “Bing (Microsoft)” and “Shopify”; ranking is “based on a number of factors” with “no way to guarantee top placement”: OpenAI Help Center, “ChatGPT Search”.
- “GPTBot is used to crawl content that may be used in training”; “OAI-SearchBot is used to surface websites in search results”; opting out of OAI-SearchBot means a site “will not be shown in ChatGPT search answers”; ChatGPT-User is “not used to determine whether content may appear in Search”: OpenAI, “Overview of OpenAI Crawlers”.
- “Bing powers a wide range of AI and search experiences, including Yahoo, DuckDuckGo, Microsoft Start, and AI tools like ChatGPT and Inflection.ai”: Microsoft Bing Webmaster Blog, June 17, 2025.
- SearchGPT prototype, the origin of ChatGPT search: OpenAI, “SearchGPT Prototype”, July 25, 2024.
- Bing Search APIs “retired on August 11, 2025”; migrate to “Grounding with Bing Search as part of Azure AI Agents”: Microsoft Learn, “Bing Search APIs Retiring on August 11, 2025”, announced May 15, 2025.
- Statistics, quotations, and citations measurably raise AI visibility: Aggarwal et al., “GEO: Generative Engine Optimization”, arXiv:2311.09735 (KDD 2024).
- ChatGPT and Perplexity share 11.0% of cited sources across 100,000 prompts: Profound, July 2025.
Every architecture claim quotes OpenAI's or Microsoft's own documentation, blog, or help center, retrieved and verified on July 22, 2026. OpenAI does not disclose its ranking algorithm or index; claims about ranking are limited to what its help center states. Study figures are cited as published by their sources.