Training · OpenAI
GPTBot
OpenAI's training crawler. Pages it fetches may be used to train future OpenAI models. It is the most-blocked AI agent on the web, and blocking it does not affect ChatGPT search citations, which come from OAI-SearchBot.
- Operator
- OpenAI
- Purpose
- Training crawlers
- Feeds
- ChatGPT
- robots.txt token
GPTBot- Honours robots.txt
- Honours robots.txt
- Documentation
- platform.openai.com
What it looks like in server logs
A representative user-agent string. Version numbers change; the GPTBot token does not. Paste your own log line into the user-agent checker to confirm a match.
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.2; +https://openai.com/gptbotHow to block GPTBot
Add this to the robots.txt at the root of your domain. Robots.txt is advisory: it works only on crawlers that choose to read it. This one is honours robots.txt according to its operator.
User-agent: GPTBot
Disallow: /
To block every training crawler at once, or to block training while keeping AI search citations, use the robots.txt generator.
How to allow GPTBot while blocking others
A more specific User-agent group wins over a wildcard. Place this above any User-agent: * block that disallows paths, and GPTBot keeps full access.
User-agent: GPTBot
Allow: /
How to verify a genuine hit
The user-agent string is self-asserted: anyone can send it from a laptop, and scrapers routinely impersonate well-known crawlers to slip past filters aimed at unknown ones.
OpenAI publishes its IP ranges as JSON; a hit from outside them is spoofed.