ChatGPT agent

What ChatGPT agent does

A browser agent drives a real browser to get something done for a person: compare, shop, book or sign up. It loads your pages, clicks and fills in forms the way a person would, so it can become a customer or give up halfway.

How to recognize it

ChatGPT agent browses like an ordinary Chrome browser, so its user agent names no agent. What sets it apart is that it signs its requests, and names https://chatgpt.com in the Signature-Agent header. A person's own browser never signs its requests.

A header is only a claim, though: anyone can send it. What proves the request is the signature, checked against the keys OpenAI publishes.

How to tell it's really ChatGPT agent

It signs its requests with Web Bot Auth (HTTP Message Signatures, RFC 9421): each request carries a signature made with an Ed25519 key, and a Signature-Agent header naming https://chatgpt.com, where the public keys are published. A valid signature verifies the request on its own, wherever it comes from.

Ethogram runs these checks on every request and gives each visit a verdict: verified, unverified or spoofed. Spoofed visits are counted apart and never added to ChatGPT agent's numbers. How verdicts work.

Can you block it in robots.txt?

Not by name. robots.txt groups its rules by user-agent token, and ChatGPT agent sends none of its own, so only the rules for every agent (User-agent: *) can apply to it. To turn it away, use your firewall or bot protection. Before you do, remember it may be doing a job for a person, such as buying something.

See what ChatGPT agent does on your site

Most analytics filter ChatGPT agent out as a bot. Ethogram records it on your server, where every request lands, with no script to load, and shows:

• Every visit from ChatGPT agent, with a verdict on whether it's really ChatGPT agent.
• Each visit step by step: the pages it read, what it searched for, and where it was blocked, looped or hit an error.
• How often its visits reach your goals, such as a purchase or a sign-up, ranked against other agents.
• Impostors that use its name, caught and counted apart from the real ChatGPT agent.
• A heads-up when its visits spike or stop, or when it starts getting blocked.