How to allow, verify or block ChatGPT cloud browser (dots, ChatGPT Work)
Verify the RFC 9421 signature against the keys OpenAI publishes at chatgpt.com/.well-known/http-message-signatures-directory and let passing requests through. On Cloudflare, a WAF skip rule on detection ID 129220581 does the same job.
Checked October 20, 2026. OpenAI's allowlisting guide covers ChatGPT's cloud browser in general and does not name dots, which has no token of its own.
How it identifies itself
| Signal | Value | Documented by the maker |
|---|---|---|
| web-bot-auth | Signature-Agent: "https://chatgpt.com" (with Signature and Signature-Input headers, RFC 9421) | Yes help.openai.com |
| header | Cloudflare Bot Detection ID 129220581 (bot tag chatgpt-agent) | Yes help.openai.com |
| user-agent | ChatGPT-User/1.0 (+https://openai.com/bot) | Yes developers.openai.com |
| ip-list | https://openai.com/chatgpt-user.json | Yes developers.openai.com |
- Allow it
- Verify the RFC 9421 signature against the keys OpenAI publishes at chatgpt.com/.well-known/http-message-signatures-directory and let passing requests through. On Cloudflare, a WAF skip rule on detection ID 129220581 does the same job.
- Verify it
- A signed request carries Signature-Agent: "https://chatgpt.com" plus Signature and Signature-Input headers that must validate against OpenAI's key directory. For unsigned fetches, match the ChatGPT-User token and check the source IP against openai.com/chatgpt-user.json.
- Block it
- Block the Cloudflare detection ID or the ChatGPT-User IP ranges at your firewall. OpenAI says robots.txt rules may not apply to ChatGPT-User because a person started the request, so a robots.txt line alone is advisory here.
robots.txt lines for ChatGPT cloud browser (dots, ChatGPT Work)
Copy the group you want for ChatGPT-User. The most specific matching group wins over your * group.
Let it in
User-agent: ChatGPT-User
Allow: /Keep it out
User-agent: ChatGPT-User
Disallow: /What we can and can't confirm
Every signal above is published by the maker. Even so, a user agent string is easy to fake, so check requests against the maker's IP list or signature where one exists before you treat them as genuine.