What Is ChatGPT Search?
ChatGPT search is ChatGPT answering with live web results and links to the pages it used. OpenAI does not publish how it picks sources. Studies suggest it runs a few related searches, then cites pages whose titles closely match them. Only 14.4% of free, logged-out responses searched at all in one study 2. To be eligible, allow the OAI-SearchBot crawler 1.
- ChatGPT search means ChatGPT looks things up on the web and cites links. Ahrefs found only 14.4% of free, logged-out answers did this 2.
- OpenAI says OAI-SearchBot is the search crawler. Sites that opt out will not be shown in ChatGPT search answers 1.
- Ahrefs found cited titles matched the prompt more closely than uncited ones (0.602 against 0.484 similarity), in one study with unclear dates, so treat it as direction 3.
- Across four assistants, about 12% of cited links were in Google's top 10 for the same prompt. For ChatGPT's in-text citations it was 8.0% 5.
- In our test, two AI tools listed 15 sites for this question, and no site appeared on both lists.
What is ChatGPT search?
ChatGPT search is the mode where ChatGPT looks up pages on the web and answers with links to the sources it used. Without a lookup, ChatGPT answers from what the model learned in training. This search integration is built into the chat, so you do not open a separate search engine.
How often does the lookup happen? Ahrefs checked 66.6 million free, logged-out ChatGPT responses from August and September 2026. Only 14.4% contained fan-out queries, which Ahrefs treats as the sign that a search ran 2. That is roughly one answer in seven.
A fan-out query is one of the extra searches an AI model runs behind your question. ChatGPT ran about 1.6 of them for each response that searched 2. Our page on query fan-out explains the idea in full.
Treat the 14.4% as a snapshot. It covers one model (GPT-5.6 Luna) on the free tier, and Ahrefs warns that OpenAI changes retrieval often. A separate API study by Nectiv, using a different model (GPT-5.6 Sol) on about 4,000 prompts, recorded 7.61 fan-outs per prompt. The two figures measure different things, so they are not directly comparable 2.
How does ChatGPT choose which websites to cite?
OpenAI publishes no ranking formula for ChatGPT search, so everything in this section comes from outside studies. Ahrefs analysed 1.4 million ChatGPT prompts and found ChatGPT cited about half (49.98%) of the URLs it retrieved 3. Finding a page and citing it are two separate steps.
Title match stood out. Cited titles scored 0.602 on similarity to the user's prompt, against 0.484 for retrieved but uncited titles. The best match to a fan-out query averaged 0.656 for cited titles 3. In plain terms, a title that reads like the question gets cited more often.
Two smaller signals appeared. Search results with natural-language URL slugs were cited 89.78% of the time, against 81.11% for those without 3. Cited search pages had a median age of about 500 days, so old pages were not shut out 3.
These are correlations from one model version. Ahrefs used open-source embeddings as a stand-in for ChatGPT's own scoring, and the article's collection date is unclear: it says February 2025 for a model it names 5.2 3. Treat the numbers as direction, not as a recipe.
| Source type | Cited when retrieved | What it suggests |
|---|---|---|
| Regular search results | 88.46% | Ordinary web pages are cited most of the time they are found |
| News | 12.01% | Many news pages are read but not credited |
| 1.93% | Heavily retrieved, rarely credited | |
| YouTube | 0.51% | Rarely cited when retrieved in this sample |
| Academic sources | 0.40% | Rarely cited in this sample |
Which OpenAI crawlers matter for ChatGPT search?
One crawler decides search inclusion: OAI-SearchBot. OpenAI says it is the crawler for search, and sites that opt out will not be shown in ChatGPT search answers, though they can still appear as navigational links 1.
The other two bots do different jobs, and OpenAI says each robots.txt setting is independent of the others 1. GPTBot relates to training: disallowing it signals that content should not be used to train OpenAI's foundation models. ChatGPT-User handles actions a person starts, is not used to crawl automatically, and robots.txt rules may not apply to it 1.
Changes are not instant. OpenAI says it takes about 24 hours from a robots.txt update for its systems to adjust 1. That is the delay for a rule change, not a forecast of when new pages get cited.
Readability matters as well. Vercel's late-2024 analysis of its network found that none of the major AI crawlers it tested, including OAI-SearchBot, rendered JavaScript 8. That data is old, so test your own pages with the free AI Crawler Checker. The comparison page on GPTBot vs OAI-SearchBot vs ChatGPT-User goes deeper.
- User-agent: OAI-SearchBot
- Allow: /
- User-agent: GPTBot
- Disallow: /
| Bot | Its job | Effect of disallowing it |
|---|---|---|
| OAI-SearchBot | Surfaces sites in ChatGPT search | Site not shown in search answers; may still appear as a navigational link |
| GPTBot | Collects content that may train models | Signals content should not be used for training |
| ChatGPT-User | Visits pages for user-started actions | robots.txt rules may not apply, so do not rely on it |
Do you need to rank on Google to be cited by ChatGPT?
No, and the overlap is small. For ChatGPT's in-text citations, only 8.0% of cited links appeared in Google's top 10 for the same prompt. Ahrefs tested 15,000 long-tail prompts, and averaged across ChatGPT, Gemini, Copilot and Perplexity the figure was about 12% 5.
Perplexity, at 28.6%, lifts that average; Gemini and Copilot were each about 8% 5. Ahrefs says about 80% of AI-cited URLs did not rank anywhere in Google for the original prompt. It offers query fan-out as the reason, which is a theory, not a tested result 5.
The ChatGPT data was collected in July 2025 and covers long-tail prompts only, so it may have shifted since 5. Our reading: aim at the narrow sub-questions around your topic, not only the headline phrase. A focused page can be found by a related search even when it ranks nowhere for the main one. That is an inference.
Which websites does ChatGPT cite most?
In Semrush's tracking of more than 230,000 prompts over 13 weeks to 12 October 2025, Reddit and Wikipedia were ChatGPT's two most-cited domains 4. Medium, Forbes and LinkedIn also appeared often.
The pattern is unstable. Reddit was cited in close to 60% of ChatGPT responses in early August 2025 and around 10% by mid-September. Wikipedia fell from roughly 55% to under 20% 4. Semrush says its data shows how citations moved, not why.
Citing is not the same as retrieving. In Ahrefs' data Reddit made up 67.8% of the retrieved but uncited URLs 3. A separate Semrush study of 217,000 prompts reported Reddit links in about 13% of ChatGPT Search responses (12.6% in its summary); the article does not define the base clearly 6. Both firms sell AI visibility tools, so read their numbers as indicators.
The lesson for a small business is simple. Do not build a plan on one site's share this month. Build pages that answer your buyers' questions, and earn mentions in several places.
What did two AI tools use as sources for this question?
On 20 October 2026 we gave this page's question to Gemini 2.5 Flash with Google Search grounding and to a Claude agent with a web search tool. We did not run ChatGPT itself. Each tool ran 3 searches, and together they listed 15 sites with none on both lists 7. The Claude agent was told to research thoroughly and Gemini chose its own number of searches, so the counts do not compare the tools. The recorded runs are held in Chalk Labs' research log.
Only the Claude run listed OpenAI's own documentation (developers.openai.com), and its third search asked about the crawler directly. The Gemini list held no OpenAI domain. This is why we cite OpenAI's documents ourselves. Two runs on one day illustrate the pattern and prove nothing about it.
| Tool | Search it ran | What it seems to be after |
|---|---|---|
| Gemini | how does ChatGPT choose websites to cite | The selection question |
| Gemini | ChatGPT citation mechanism | The technical mechanism |
| Gemini | ChatGPT search integration and citations | How the product combines search and answers |
| Claude | ChatGPT search how it picks sources to cite | The selection question, worded differently |
| Claude | ChatGPT citations most cited domains study | Data on which domains win |
| Claude | OAI-SearchBot robots.txt ChatGPT search inclusion | The access rule for search inclusion |
What do people get wrong about ChatGPT search?
'Blocking GPTBot removes me from ChatGPT search.' OpenAI treats its robots.txt settings as independent, and ties search inclusion to OAI-SearchBot 1. Check which bot your rule names.
'ChatGPT always searches the web.' In the Ahrefs sample only 14.4% of responses did 2. Many answers never touch your site.
'Ranking first on Google means a citation.' Overlap with Google's top 10 was only 8.0% for ChatGPT's in-text citations 5.
'The same sites are always cited.' Reddit and Wikipedia swung sharply within weeks in 2025 4. Re-check your own results monthly.
What can you do about ChatGPT search before Friday?
Start with access, then wording. None of these steps costs money, and you can finish them in an afternoon.
- Open yourdomain.com/robots.txt and confirm OAI-SearchBot is not disallowed, directly or through a blanket rule 1.
- Run the AI Crawler Checker and view a key page with JavaScript off. If the text is missing, fix rendering first.
- Rewrite page titles so they read like the question a buyer would type, because title match tracked with citation 3.
- Use readable URL slugs such as /how-much-does-a-roof-repair-cost, since natural-language slugs were cited more often 3.
- Ask ChatGPT your ten most important buyer questions. Note whether it searched, which sites it cited, and whether you appear.
- Look for the extra searches behind each question with the free Query Fan-Out Explorer and cover the gaps. For a full plan, read how to get recommended by ChatGPT.
What an AI assistant searches when it answers this
We gave the question below to AI search tools and recorded the web searches they ran behind the scenes. This is called query fan-out. Each phrase is a page you could publish to be found and cited.
“how does ChatGPT search choose which websites to cite”
The searches it ran- ChatGPT search how it picks sources to cite
- ChatGPT citations most cited domains study
- OAI-SearchBot robots.txt ChatGPT search inclusion
- how does ChatGPT choose websites to cite
- ChatGPT citation mechanism
- ChatGPT search integration and citations
Frequently asked questions
How does ChatGPT choose which websites to cite?
OpenAI does not publish the method. Ahrefs found ChatGPT cited about half of the URLs it retrieved, and cited titles matched the prompt and fan-out queries more closely than uncited ones 3. Treat that as a pattern from one model version, not a rule.
Do I need to allow OAI-SearchBot?
Yes, if you want to appear in ChatGPT search answers. OpenAI says sites that opt out of OAI-SearchBot will not be shown there, though they can still appear as navigational links. It also says a robots.txt change takes about 24 hours to apply 1.
Does blocking GPTBot hide my site from ChatGPT search?
OpenAI says each robots.txt setting is independent. GPTBot relates to training, while OAI-SearchBot relates to search 1. Disallowing GPTBot alone should not remove you from search answers, but check your rules for a blanket block that catches every bot.
How often does ChatGPT search the web?
Less often than many people assume. Ahrefs found 14.4% of 66.6 million free, logged-out responses contained fan-out queries in August and September 2026 2. Paid and reasoning use may differ, and commercial 'best X' prompts triggered more searches in other studies.
Do I have to rank in Google's top 10 to be cited?
No. Ahrefs found 8.0% of ChatGPT's in-text citations appeared in Google's top 10 for the same prompt. Averaged across four assistants, about 12% did 5. Ranking still helps your site get found, but it is not the only route.
Sources
- Overview of OpenAI Crawlers · OpenAIOpened October 20, 2026. Supports: OAI-SearchBot is for search and opted-out sites are not shown in ChatGPT search answers but can appear as navigational links; GPTBot and training; ChatGPT-User is user-initiated and robots.txt may not apply; each setting is independent; about 24 hours for robots.txt changes.
- How Often Does ChatGPT Search the Web? Query Fan-Out Study · AhrefsOpened October 20, 2026. Supports: Published 12 October 2026: 14.4% of 66.6 million free, logged-out GPT-5.6 Luna responses contained fan-out queries; about 1.6 per response; caveats about one model and commercial prompts (7.61 per prompt in one API study).
- Why ChatGPT Cites One Page Over Another · AhrefsOpened October 20, 2026. Supports: 1.4 million prompts: 49.98% of retrieved URLs cited; citation rates by source type; title similarity 0.602 vs 0.484 and 0.656; natural-language slugs 89.78% vs 81.11%; median page age about 500 days; Reddit 67.8% of retrieved but uncited; caveats on dates and method.
- The Most-Cited Domains in AI: A 3-Month Study · SemrushOpened October 20, 2026. Supports: Published 10 November 2025; 230,000+ prompts over 13 weeks to 12 October 2025; Reddit and Wikipedia top ChatGPT domains; Reddit fell from close to 60% to around 10% of responses and Wikipedia from roughly 55% to under 20% between August and September 2025.
- Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt · AhrefsOpened October 20, 2026. Supports: Published 11 August 2025; 15,000 long-tail prompts; about 12% average overlap across ChatGPT, Gemini, Copilot and Perplexity (Perplexity 28.6% lifts the average); ChatGPT in-text 8.0%, Gemini 8.6%, Copilot 8.2%; about 80% of cited URLs not ranking in Google; ChatGPT data collected July 2025.
- Reddit Visibility in AI Search · SemrushOpened October 20, 2026. Supports: Published 10 November 2025; about 217,000 prompts; Reddit about 13% of ChatGPT Search responses (12.6% in key takeaways), 9% of AI Mode, 4% of Perplexity.
- Query Fan-Out Explorer (observed examples) · Chalk LabsOpened October 20, 2026. Supports: From Chalk Labs' research log: the searches recorded on 20 October 2026 for the prompt 'how does ChatGPT search choose which websites to cite': Gemini 2.5 Flash with Google Search grounding (3 searches) and a Claude agent with a web search tool (3 searches), and the sites each run listed.
- The rise of the AI crawler · VercelOpened October 20, 2026. Supports: Dated 17 December 2024: none of the major AI crawlers tested, including OAI-SearchBot, rendered JavaScript.
- How Often Do AI Assistants Hallucinate Links? · AhrefsOpened October 20, 2026. Supports: Published 2 September 2025: ChatGPT cited 404 URLs 2.38% of the time against 0.84% for the top 20 Google results in the second test.