GPTBot trains OpenAI's models. OAI-SearchBot builds the index that ChatGPT search cites and links. ChatGPT-User loads a page live because someone just asked ChatGPT about it. If you want ChatGPT to recommend you, keep OAI-SearchBot open. Blocking GPTBot only keeps your future pages out of training, and blocking ChatGPT-User may not work at all, because OpenAI says robots.txt may not apply to it.
The figures are site-weighted means across the 12 small-business sites in our Q4 2026 Observatory, August to October 2026. Every fact about what the bots do comes from OpenAI's own crawler documentation, checked in October 2026.
The three bots side by side
| GPTBot | OAI-SearchBot | ChatGPT-User | |
|---|---|---|---|
| Job | Collects pages that may train OpenAI's models | Builds the index ChatGPT search uses to show and link sites | Fetches one page live when a ChatGPT user or a Custom GPT asks about it |
| robots.txt token | GPTBot |
OAI-SearchBot |
ChatGPT-User |
| Follows robots.txt | Yes | Yes; OpenAI says changes take about 24 hours | May not: OpenAI says rules may not apply to user-initiated visits |
| Block it and you… | stay out of future training data, with no effect on ChatGPT search | drop out of ChatGPT search results and their citations | stop ChatGPT reading the page for that user, if the rule is honoured |
| Verify with | openai.com/gptbot.json | openai.com/searchbot.json | openai.com/chatgpt-user.json |
| Share of AI fetches in our data | 10.1% | 14.9% | 28.1% |
The user agents carry a version number (GPTBot/1.4 and OAI-SearchBot/1.4 in OpenAI's docs as of October 2026, ChatGPT-User/1.0), which OpenAI says can change. Match on the product token, such as GPTBot, not on the full string.
Which one sends you customers
None of them sends visitors directly. What brings people to your site is a ChatGPT answer that links to you, and that depends on OAI-SearchBot having your page in its index. When someone then asks a follow-up about your business, ChatGPT-User is often the agent that fetches the page to answer it. That is why ChatGPT-User was the busiest AI agent in our data: on the typical site, just over a third (35.1%) of AI fetches were live lookups triggered by a person, not background crawling.
The gap between reading and sending is wide. Across the same 12 sites, AI crawlers made about 120 fetches for every visit an AI assistant referred (details). To see how many of your own visits come from ChatGPT, follow our GA4 guide to ChatGPT traffic.
Block training, keep ChatGPT search
This is the most common choice we see, and it costs nothing in ChatGPT search:
# Keep future pages out of OpenAI training
User-agent: GPTBot
Disallow: /
# ChatGPT search and live fetches stay allowed
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
OpenAI has also said it will honour the new shared __usage_ai-training robots.txt group, but hasn't given a date, so keep the GPTBot group for now (details).
The same split works for other operators: Anthropic has ClaudeBot (training), Claude-SearchBot (search) and Claude-User (live fetch), and Perplexity has PerplexityBot and Perplexity-User. Our robots.txt guide has full recipes for every operator, and the AI crawler registry lists 32 AI agents and robots.txt tokens with their documentation.
Mistakes to avoid
- Blocking GPTBot to "get out of ChatGPT". It doesn't. OAI-SearchBot is the one ChatGPT search depends on.
- Blocking OAI-SearchBot by accident. A blanket
User-agent: */Disallow: /, or a firewall rule aimed at "AI bots", removes you from ChatGPT search. Check with our AI access checker. - Trusting the user agent. Anyone can send "GPTBot" in a request. If you allow or block by firewall, use OpenAI's published IP lists.
- Expecting an instant change. OpenAI says OAI-SearchBot takes about 24 hours to pick up a robots.txt change.
Check your own site
Run your domain through the AI access checker. It reads your live robots.txt and reports, bot by bot, whether GPTBot, OAI-SearchBot, ChatGPT-User and the other major agents can reach your pages.
Questions people ask
What is the difference between GPTBot and OAI-SearchBot?
GPTBot collects pages that may be used to train OpenAI's models. OAI-SearchBot builds the index ChatGPT search uses to show and link websites. They are controlled separately in robots.txt, so you can block one and allow the other.
Does blocking GPTBot remove my site from ChatGPT?
No. OpenAI says GPTBot is for training and that ChatGPT search uses OAI-SearchBot. Block GPTBot and allow OAI-SearchBot, and ChatGPT can still find and cite your pages.
Does ChatGPT-User follow robots.txt?
Not necessarily. OpenAI says robots.txt rules may not apply to ChatGPT-User, because a person asked ChatGPT to look at the page. OpenAI also says ChatGPT-User does not decide whether a site appears in ChatGPT search.
Which OpenAI bot visits small-business sites most?
ChatGPT-User. On the 12 small-business sites in our Q4 2026 Observatory it was 28.1% of the typical site's AI fetches, against 14.9% for OAI-SearchBot and 10.1% for GPTBot.
How do I check a request is really from OpenAI?
Match the source IP against the list OpenAI publishes for that bot: openai.com/gptbot.json, openai.com/searchbot.json and openai.com/chatgpt-user.json. The user agent string alone can be faked.