ChatGPT-User: What robots.txt Can and Cannot Control

ChatGPT-User is not the same thing as GPTBot or OAI-SearchBot. It can appear when a ChatGPT user asks the product to visit or retrieve a web page. Because the request originates from a user action, OpenAI says robots.txt rules may not apply.

What ChatGPT-User is for

ChatGPT-User supports certain user actions in ChatGPT and Custom GPTs. It is not used for automatic crawling of the web, and OpenAI says it is not the control used to decide whether a site can appear in ChatGPT Search.

That distinction matters because three different goals require three different controls:

• For training crawling, review GPTBot.
• For ChatGPT Search visibility, review OAI-SearchBot.
• For user-triggered retrieval, the relevant user agent is ChatGPT-User, but robots.txt should not be treated as a hard security boundary.

Should you add a ChatGPT-User group?

You can publish a group that expresses your preference, but do not assume it guarantees that every user-triggered request will be blocked. If a resource must not be fetched by unauthenticated clients, protect it with authentication, authorization, server rules, or a WAF rather than robots.txt.

User-agent: ChatGPT-User
Disallow: /private/

This example documents a crawl preference for a path. It is not a replacement for protecting private content.

If you need to prevent access to private material

Do not use a ChatGPT-User robots.txt group as an access-control mechanism. OpenAI states that user-initiated requests may not follow robots.txt restrictions. For private endpoints, require authentication and enforce authorization at the server or application layer. If rate limiting or bot challenges are needed, configure them separately and verify that ordinary legitimate users are not blocked.

If your real goal is reducing automated training crawling, test GPTBot rules instead. If your concern is ChatGPT search discovery, test OAI-SearchBot instead. You can inspect the hypothetical policy with the draft Checker, but that simulator cannot prove how a user-triggered retrieval will behave in the product.

OpenAI’s crawler documentation is the source for the distinction between automated crawlers and ChatGPT-User.

Do not block the wrong OpenAI agent

A common configuration mistake is to block ChatGPT-User when the real goal is to opt out of training, or to block GPTBot when the real goal is to control ChatGPT Search. Keep the groups separate so each rule maps to the outcome you actually want.

Use the robots.txt Checker to validate syntax and path matching. If your goal is a clean OpenAI policy with training and search handled independently, see the OpenAI robots.txt controls template.