OpenAI runs several different crawlers, and they do different jobs. GPTBot collects content that may be used for training. OAI-SearchBot reads pages for ChatGPT's search results. ChatGPT-User opens a page because a person asked ChatGPT something. Treating them as one "ChatGPT bot" leads to the wrong robots.txt rules and the wrong conclusions.
The short version
| Agent | What it is for | Triggered by | robots.txt |
|---|---|---|---|
| GPTBot | Crawling content that may be used to train OpenAI's models | OpenAI's crawl schedule | Applies. Disallowing it signals your content should not be used for training. |
| OAI-SearchBot | Surfacing websites in ChatGPT's search features | OpenAI's crawl schedule | Applies. Opting out removes you from ChatGPT search answers. |
| ChatGPT-User | Visiting a page when a user asks ChatGPT or a Custom GPT a question | A person's request | OpenAI says the rules may not apply, because the user started the action. |
| OAI-AdsBot | Checking pages submitted as ads on ChatGPT | An ad submission | Not stated by OpenAI. |
These descriptions follow OpenAI's own crawler documentation, checked on November 15, 2026. Operators change these pages, so read the source before changing your rules.
GPTBot: training
GPTBot crawls content that may be used in training OpenAI's generative models. It sends you no visitors. Whether to allow it is a business choice: some publishers want their material represented in models, others do not. A Disallow rule for GPTBot in robots.txt is the documented way to say no.
OAI-SearchBot: ChatGPT search
OAI-SearchBot is for search. OpenAI says it is used to surface websites in ChatGPT's search results. If you want to be quoted or linked in ChatGPT search answers, this is the crawler to allow. According to OpenAI, a site that opts out will not be shown in ChatGPT search answers, though it can still appear as a navigational link.
A common mistake is blocking GPTBot and OAI-SearchBot together when the intent was only to opt out of training. They are controlled separately.
ChatGPT-User: a person's question
ChatGPT-User appears when someone asks ChatGPT or a Custom GPT something and it visits a page to answer. OpenAI says it is not used for crawling the web automatically and is not used to decide whether content appears in search. Because a person started the action, OpenAI says robots.txt rules may not apply.
For a website owner this is the most interesting of the three. Each ChatGPT-User request follows a real question from a real person. You cannot see the question, but you can see which pages get fetched, and how that changes over time.
Example robots.txt rules
To opt out of training while staying available to ChatGPT search:
User-agent: GPTBot
Disallow: /
User-agent: OAI-SearchBot
Allow: /robots.txt is a request, not a lock. Well-behaved crawlers follow it; software pretending to be one does not have to.
Is it really OpenAI?
Anyone can send a request with "GPTBot" in its user agent. OpenAI publishes the network ranges each agent uses (gptbot.json, searchbot.json and chatgpt-user.json), so a request can be checked against them. A request that claims the name from an unlisted address is a claim that failed its check. How to verify AI crawlers walks through it.
Seeing the three in your own traffic
None of these normally appears in Google Analytics, because they do not run its tag. They appear in server-side records. AI crawler analytics explains what to measure. PulseAI Analytics names each of these agents separately, checks them against OpenAI's published networks, and groups them by job, so training crawling, search crawling and people's questions are three numbers instead of one. It reports and does not block. Installation guide.
Quick answers
What is the difference between GPTBot and ChatGPT-User?
GPTBot crawls on a schedule to collect content that may be used for training. ChatGPT-User visits a page only when a person asks ChatGPT or a Custom GPT something. The first is automatic crawling; the second follows a real question.
If I block GPTBot, will I disappear from ChatGPT search?
Not by that rule alone. ChatGPT search uses OAI-SearchBot, which is controlled separately in robots.txt. Blocking GPTBot signals that your content should not be used for training.
Does robots.txt stop ChatGPT-User?
OpenAI says robots.txt rules may not apply to ChatGPT-User, because the visit is started by a user rather than by automatic crawling.
How can I tell if a GPTBot request is real?
Check the request's address against the network ranges OpenAI publishes for GPTBot. A user agent alone is only a claim, and anyone can send it.