ChatGPT Search: How It Works, OAI-SearchBot and Citations

ChatGPT search is ChatGPT’s web search: it answers a question with current information and links to the sources it drew on. ChatGPT may run it on its own when a question needs fresh information, or you can choose Search. To be eligible to appear, a site must let OAI-SearchBot crawl it.

This guide covers what ChatGPT search is, how it finds and shows sources, what OAI-SearchBot does and how to allow it in robots.txt, and how to check whether ChatGPT cites your pages. Every statement about OpenAI’s products is taken from OpenAI’s own pages and linked.

Last checked 5 November 2026 against OpenAI’s own pages.

What is ChatGPT search?

OpenAI describes it as ChatGPT searching the web to answer questions “with current information and links to relevant sources”. ChatGPT may search automatically when a question would benefit from current information, and a person can start a search by hand from the tools menu or by typing / in the message box. (source)

• Who can use it. OpenAI lists web search on the Free, Go, Plus, Pro, Business, Enterprise and Edu plans, and says people who are not signed in can use it too. Searches count against a plan’s usage limits. (source)
• Where. On chatgpt.com and in the ChatGPT desktop and mobile apps. OpenAI also offers a Chrome extension that makes ChatGPT the browser’s default search engine. (source)
• Since when. OpenAI introduced ChatGPT search on 31 October 2024, and on 5 February 2025 made it available to everyone in regions where ChatGPT is available, with no signup required. (source)

ChatGPT search is not the sidebar search that finds your own past chats, projects and files. OpenAI treats those as separate features. (source)

How does ChatGPT search pick its sources?

OpenAI describes three parts: the queries ChatGPT sends, the providers it sends them to, and how it ranks what comes back.

It rewrites the question into shorter queries

When ChatGPT search uses other search providers, it “typically rewrites your query into one or more targeted queries”. OpenAI’s example turns a researcher’s long question about cancer drugs that target CCR8 into a short query naming the drug class and the year, then sends narrower follow-up queries after reading the first results. (source)

Two things can change the rewritten query. ChatGPT may add a general location estimated from the IP address (“top restaurants San Francisco” for a question about restaurants nearby), and, when memory is on, it may use saved memories, such as a stated diet or city. (source) For a brand this means your pages compete for the short query ChatGPT writes, which can differ from person to person, not for the full question someone typed.

It sends the queries to search providers

OpenAI says ChatGPT search “sometimes partners with other search providers”, and the privacy statements it links for them are Microsoft’s and Shopify’s. (source) At launch it said ChatGPT search uses third-party search providers as well as content provided directly by its partners. (source) OpenAI does not say how much weight each provider’s results carry.

It ranks the results on factors OpenAI does not list

OpenAI’s wording is that ChatGPT “ranks search results using multiple factors intended to help users find relevant, reliable information”, and that placement is not guaranteed. (source) The factors are not published. Anyone offering a guaranteed place in ChatGPT search is promising something OpenAI itself does not.

For how answer engines in general choose which passage to quote, see how AI engines choose citations. For the content work that follows from it, see how to rank in ChatGPT.

How does ChatGPT search show sources?

OpenAI describes three places where a source can appear in an answer that used search. (source)

• Citations in the answer. Responses that use web search may include citations; selecting one opens the source, and on desktop web pointing at it shows a preview.
• The Sources list. Selecting Sources, when available, shows “cited sources and other relevant links”.
• Images and maps. An image in a response links to its source, and on iOS and Android results may include a map.

The Sources wording matters for measurement. A page can appear there as a relevant link without being cited in the answer text, so record the two separately. OpenAI also tells users that search results and citations “can be incomplete, outdated, or incorrect” and to check when a cited source was published or updated, which is one more reason to keep dates and sources visible on your own pages. (source)

A citation is also not the same as being named. Your brand can be recommended in the answer while the sources are other sites, or your page can be cited for a fact while a competitor is recommended. Citation vs brand mention explains why the two are tracked as separate measures.

What is OAI-SearchBot?

OAI-SearchBot is the crawler OpenAI uses “to surface websites in search results in ChatGPT’s search features”. OpenAI says sites that opt out of it “will not be shown in ChatGPT search answers”, recommends allowing it in robots.txt, and publishes its IP addresses at openai.com/searchbot.json. Its user-agent string contains OAI-SearchBot and links to openai.com/searchbot, and when it fetches robots.txt it may add a robots.txt marker to that string. (source)

The Help Center gives the same rule from the other side: to make a site eligible for ChatGPT search, allow OAI-SearchBot to crawl it and confirm that the host or CDN allows traffic from OpenAI’s published search bot IP addresses. (source)

OpenAI runs four bots a site owner is likely to see, and says each one’s setting is independent of the others. (source)

One detail is easy to miss: if a site allows both OAI-SearchBot and GPTBot, OpenAI says it may use the results of one crawl for both purposes. (source) The free AI crawler checker reads your robots.txt for every OpenAI bot and tells you which ones it blocks, and whether a block costs you answers or only keeps you out of training data.

How to allow OAI-SearchBot in robots.txt and check you are cited

Step 1. Read your robots.txt as OAI-SearchBot would

Open your site's /robots.txt and find the group that names OAI-SearchBot. If there is none, the group for every crawler (User-agent: *) applies, so a site-wide Disallow there also keeps OAI-SearchBot out.

Under the robots.txt standard, a crawler follows the group that names it and falls back to the User-agent: * group only when none does. (source) So a site that never mentions OpenAI can still block OAI-SearchBot through a blanket rule written for other crawlers.

Step 2. Add an allow group for OAI-SearchBot

Add a group that starts with User-agent: OAI-SearchBot followed by Allow: /, or by the paths you want ChatGPT search to show. OpenAI recommends allowing OAI-SearchBot and says sites that opt out are not shown in ChatGPT search answers.

A minimal group that lets OAI-SearchBot read the whole site looks like this:

User-agent: OAI-SearchBot
Allow: /

To keep a private area out, add a Disallow line for that path inside the same group. OpenAI does not publish any other robots.txt syntax for ChatGPT search, so standard rules are all that is needed.

Step 3. Set GPTBot separately

GPTBot crawls content that may be used to train OpenAI's models. OpenAI says each bot's setting is independent, so you can disallow GPTBot and still allow OAI-SearchBot.

A site that wants to appear in ChatGPT search but stay out of training can say so in two groups, which OpenAI treats as independent settings. (source)

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

That is a business decision, not a ranking one: OpenAI does not say that allowing GPTBot helps a site appear in search.

Step 4. Let OpenAI's search bot addresses through your CDN and firewall

OpenAI publishes the IP addresses OAI-SearchBot uses at openai.com/searchbot.json. Check that your host, CDN or firewall lets them through, because a block there stops the crawler whatever robots.txt says.

Bot-management and firewall rules often sit in front of the site, so a block there does not show in the robots.txt file at all. Allow the addresses on OpenAI’s published list rather than every request that claims the name, since a user-agent string is easy to copy. (source)

Step 5. Allow about 24 hours

OpenAI says it can take about 24 hours from a robots.txt update for its search systems to adjust.

That is the only timing OpenAI publishes. (source) Being crawled again does not guarantee a citation; it only makes the site eligible.

Step 6. Confirm the visits in your logs

Look for the OAI-SearchBot user agent in your server or CDN logs, and check that the requests come from addresses on OpenAI's published list.

Requests from ChatGPT-User in the same logs mean a person asked ChatGPT about a page, which is a different signal from OAI-SearchBot’s crawling. If OpenAI’s bots are blocked by mistake, ChatGPT agent blocked your site? covers how to find the cause.

Step 7. Check whether answers cite you

Ask ChatGPT your buyers' questions with Search on, open Sources and record whether your pages are cited, on a fixed schedule rather than once.

• Use your buyers’ questions. Ask what a buyer would ask, including questions that do not name you. Which prompts to track explains how to pick them.
• Sign out, or turn memory off. OpenAI says memory and location can shape the query ChatGPT sends, and that signed-out users can search, so a signed-out session is closer to what a new buyer sees. (source)
• Record three things per answer. Whether your brand is named, whether your page is cited in the text, and whether it appears in Sources.
• Repeat. One run tells you little, because answers and their sources change from run to run. How to track brand mentions in ChatGPT sets out a routine you can run by hand.

Do ChatGPT ads affect ChatGPT search results?

OpenAI says they do not: “Ads do not influence ChatGPT’s answers.” Ads may appear below a response for users on the Free and Go plans, are labelled, and run on systems separate from the chat model. (source) An ad is therefore not a citation, and the two belong in separate lines of any report. One link between them is technical: OpenAI’s ad guide says ad landing pages must not block OAI-AdsBot or OAI-SearchBot. (source) Our guide to ChatGPT ads and AI visibility covers how to report the two apart.

Where Noma fits

Noma by Nurdd is an answer engine optimisation (AEO) platform. It asks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, the questions a brand's buyers ask, every day, and shows where the brand is named, where competitors are named instead, and which pages the answers draw on.

For ChatGPT, the ChatGPT visibility tracker asks ChatGPT your prompts once a day on paid plans and records whether your brand is named, where it sits against competitors, how it is described and which URLs ChatGPT cited. Plans start at $79 a month (Starter), and the trial runs for 3 days with 25 prompts; details are on the pricing page. For a first look without an account, run a free brand audit, and check your crawler rules with the AI crawler checker.

Sources

Each read on OpenAI’s own pages on 5 November 2026, apart from the robots.txt standard.

• OpenAI Help Center: Searching the web with ChatGPT: availability, automatic and manual search, citations and Sources, query rewriting, search providers, ranking and eligibility.
• OpenAI: Overview of OpenAI crawlers: OAI-SearchBot, GPTBot, ChatGPT-User and OAI-AdsBot, independent settings and the 24-hour delay.
• OpenAI: OAI-SearchBot IP addresses (searchbot.json).
• OpenAI: Introducing ChatGPT search: launch date, availability without signup, search providers and partner content.
• OpenAI Help Center: Ads in ChatGPT and Creating ads for ChatGPT Ads: ads and answers, and landing-page crawler access.
• RFC 9309: Robots Exclusion Protocol: how a crawler picks its robots.txt group.

← Back to the blog