GUIDE

How does ChatGPT search work for websites

ChatGPT runs 1.78 searches per prompt (Ahrefs, Aug 2025) and cites 2.2 URLs per answer on average (FoundByChat, Oct 2026).

Updated October 30, 20265 min read

55.5%Citation rate for position #1 in first search (FoundByChat, Oct 2026)
63.6%Free instant mode results from labrador index (RESONEO, Jul 2026)
2.2Average URLs cited per ChatGPT answer (FoundByChat, Oct 2026)

In short

ChatGPT may search the web automatically when a question would benefit from current information. In an August 2025 Ahrefs study of 118,931 fan-out queries, ChatGPT ran 1.78 searches per prompt on average.
OpenAI uses OAI-SearchBot to surface websites in ChatGPT search, Microsoft says its Bing-based Web IQ powers ChatGPT, and OpenAI has 18 publisher content partners.
A July 2026 RESONEO study of 1,249 answers found 63.6% of free instant mode results came from OpenAI's "labrador" index, and 74.6% of paid thinking mode results from "bright".
In FoundByChat's October 2026 study of 218 answers with web search, the first result of ChatGPT's first search was cited 55.5% of the time, and ChatGPT cited 10.3% of the web results it retrieved.

Step 1: ChatGPT decides to search and rewrites the prompt

ChatGPT may search the web automatically when a question would benefit from current information. OpenAI launched ChatGPT search on October 31, 2024 using a fine-tuned GPT-4o model. As of October 2026, web search is available across Free, Go, Plus, Pro, Business, Enterprise, and Edu plans, and to logged-out users.

When ChatGPT search partners with other search providers, it typically rewrites the query into one or more targeted queries for those providers. It may share a general IP-derived location, never the IP or account details.

Step 2: ChatGPT retrieves results from its own index and third-party providers

ChatGPT search sources and observations include the following (see our guide on where ChatGPT gets its information):

OpenAI's own crawler and index: OpenAI runs OAI-SearchBot to crawl pages for search. Between May 21 and July 21, 2026, Peec AI observed ChatGPT labeling results from its internal index as "labrador". Through a discovery pilot started in July 2026, Cloudflare reported that OpenAI indexed hundreds of millions of new pages across about 1% of Cloudflare's network by October 15, 2026.
Microsoft and Bing: At launch, OpenAI's Srinivas Narayanan said: "We use a set of services and Bing is an important one." OpenAI's Enterprise and Edu docs confirm ChatGPT may share disassociated queries with Bing, and its consumer search help page links Microsoft's and Shopify's privacy policies. Microsoft states its Web IQ service, launched June 2, 2026 on the Bing index, powers Copilot and ChatGPT. On February 6, 2025, Seer Interactive found 87% of ChatGPT search citations matched Bing's top 20 organic results, versus 56% for Google.
Scraping providers: Peec AI reported in September 2026 that ChatGPT also labeled result sources as "bright", "oxylabs", and "serp" (scraping providers). In RESONEO's July 2026 study, 34.7% of free instant mode results came from "bright". On August 21, 2025, The Information reported OpenAI used SerpApi, a Google scraper, for some real-time topics. An August 20, 2025 Ahrefs study found only 6.82% of ChatGPT's results were in Google's top 10, while 83.39% did not appear in Google's results at all.
Publisher partners: OpenAI has 18 publisher partners, including AP, Axel Springer, the Financial Times, News Corp, Le Monde, Vox Media, The Atlantic, Conde Nast, Hearst, The Guardian, The Washington Post, and Reddit.

Step 3: How OAI-SearchBot and OpenAI's other crawlers access your site

To be eligible for ChatGPT search answers, allow OAI-SearchBot and OpenAI's published IP ranges; blocked sites can still appear as navigational links. OpenAI documents four crawlers as of October 2026:

Bot name	User agent string	Purpose	Effect of blocking in robots.txt
OAI-SearchBot	`OAI-SearchBot/1.4; +https://openai.com/searchbot`	Surfaces websites in ChatGPT search features	Site is not shown in search answers, though bare navigational links may still appear
GPTBot	`GPTBot/1.4; +https://openai.com/gptbot`	Crawls content that may be used to train generative AI foundation models	Opts content out of model training; does not remove site from search
ChatGPT-User	`ChatGPT-User/1.0; +https://openai.com/bot`	Handles user actions in ChatGPT and Custom GPTs	Initiated by a user, so `robots.txt` rules may not apply; does not decide search eligibility
OAI-AdsBot	`OAI-AdsBot/1.0`	Checks safety of pages submitted as ChatGPT ads	Advertisers must allow it; not used to train models

These settings are independent, though allowing both OAI-SearchBot and GPTBot lets OpenAI use one crawl for both. ChatGPT search can take about 24 hours to adjust after a robots.txt change.

Step 4: How citation rates varied by search position in a study

OpenAI states that ChatGPT ranks search results using multiple factors without guaranteed placement. In FoundByChat's October 2026 study of 218 ChatGPT answers, ChatGPT cited 2.2 URLs per answer on average. While answers can include citations and a Sources panel (which OpenAI notes can be incomplete, outdated, or incorrect), the study reported these citation rates by position in ChatGPT's first search:

Position 1: cited 55.5% of the time (95% CI 49.1% to 61.9%).
Position 2: cited 34.9% of the time.
Position 3: cited 27.5% of the time.
Positions 4 and 5: cited 14.4% of the time.
Positions 6 to 10: cited 8.1% of the time.
Positions 11 to 15: cited 2.4% of the time.
Position 16 and lower: cited 0.4% of the time.

Retrieval alone does not guarantee a citation: Reddit made up 28% of retrieved results in the study, but only 2 of 2,462 citations. See our guides to ChatGPT citations and how to show up in ChatGPT.

How to check how ChatGPT search sees your website

Test a page with the free GEO audit: Needs no account and allows 5 audits a day. It fetches your page using the OAI-SearchBot user agent, rebuilds the title and ~200-character snippet ChatGPT keeps, looks the page up in ChatGPT, and scores it out of 100 across bot protection, HTML text without JavaScript, menu or cookie text in the snippet, robots.txt rules, and noindex tags.
Monitor prompts with FoundByChat: Add your site and customer questions, and FoundByChat asks ChatGPT with web search every 2 hours as a user in the US (English) or the Netherlands (Dutch). It records whether your page was "Found", "Cited", the usual position of your pages among search results, and each indexed page's stored title and snippet. Free to start.
Check analytics for referrals: Outbound links carry utm_source=chatgpt.com. See tracking ChatGPT traffic in GA4.

Common mistakes site owners make with ChatGPT search

Blocking OAI-SearchBot when trying to block training: Disallow GPTBot to opt content out of training and allow OAI-SearchBot and OpenAI's published IP ranges for search eligibility.
Relying on robots.txt alone to hide a page: A blocked URL from a third-party provider can still show as a bare link and title. Use a crawler-readable noindex tag instead.
Letting menus or cookie banners fill the first 200 characters: Early HTML text can leak into the ~200-character snippet ChatGPT keeps.

FOUNDBYCHAT

Know the moment ChatGPT picks up your pages

We ask ChatGPT your questions every 2 hours and mail you what its index picked up or rewrote. Free to start: sign in with Google or an email link.

Start free Or run the free GEO audit

Questions

Does blocking GPTBot remove my website from ChatGPT search?

No, blocking GPTBot only opts your content out of training OpenAI's foundation models. Search eligibility is controlled separately by OAI-SearchBot. As long as OAI-SearchBot is allowed, your pages can appear in ChatGPT search answers.

Does ChatGPT search use Google or Bing?

OpenAI states that it partners with third-party search providers, naming Bing for Enterprise and Edu and linking Microsoft's privacy policy in its consumer search help docs. Microsoft also confirms its Bing-based Web IQ service powers ChatGPT. In addition, In 2026, Peec AI observed result-source labels "labrador", "bright", "oxylabs", and "serp". RESONEO measured results from "labrador" and "bright".

Why does a blocked page still show up as a link in ChatGPT?

According to OpenAI's publisher FAQ, if you block OAI-SearchBot in robots.txt, ChatGPT can still learn the URL from a third-party search provider and show it as a navigational link with a page title. To stop the URL from showing at all, allow the crawler to read the page and include a noindex directive.

How can I track when my pages enter ChatGPT's index?

You can use FoundByChat to check your customer questions in ChatGPT search every 2 hours. After the first check sets your baseline, FoundByChat emails you when a page shows up for the first time ("New in the index"), when ChatGPT stores a different title or snippet ("Snippet changed"), or when ChatGPT returns no web results three checks in a row.