GPTBot vs. OAI-SearchBot vs. ChatGPT-User: which should you block?

OpenAI documents three distinct crawler tokens, each triggered by a different thing. Treating them as one “OpenAI bot” and blocking or allowing all three together is an easy mistake: it can remove a site from ChatGPT search when the intent was only to opt out of training.

Which OpenAI crawler should I block?

To keep content out of OpenAI’s model training while staying visible in ChatGPT search, disallow GPTBot and leave OAI-SearchBot allowed. OpenAI documents the two settings as independent. ChatGPT-User only fetches pages a user asks about, and OpenAI says robots.txt rules may not apply to it.

What does each OpenAI token do?

• GPTBot — training. Crawls content that may inform future model training.
• OAI-SearchBot — search. Surfaces and links to pages in ChatGPT search results.
• ChatGPT-User — user-triggered. Fetches a specific page because a person asked ChatGPT a direct question about it.

What happens when you block each one?

• Disallow GPTBot if you don’t want your content used for OpenAI’s model training. This has no effect on whether your site can be found through ChatGPT search.
• Leave OAI-SearchBot allowed if you want your pages discoverable through ChatGPT’s search features. OpenAI says sites that disallow it are not shown in ChatGPT search answers, though they can still appear as navigational links.
• Leave ChatGPT-User allowed unless you specifically don’t want any content served in response to direct user questions inside ChatGPT — blocking it does not reduce training or search exposure, since it’s a separate, user-initiated request path. Note that OpenAI’s own documentation states “Because these actions are initiated by a user, robots.txt rules may not apply” to this token, so a Disallow for it is a declared preference, not a dependable control.

Does a GPTBot rule cover all of OpenAI?

A single User-agent: GPTBot group is sometimes assumed to cover “all of OpenAI.” It doesn’t — robots.txt groups match on the exact token named, with no fallback to a shared vendor identity. Each token needs its own explicit Disallow if you want to restrict it. CrawlPact’s AI crawler checker shows the result for all three, evaluated independently, so a gap is visible immediately.