Short answer
ChatGPT agent uses a cloud browser to read pages, click buttons and fill in forms on a person's behalf. OpenAI says it requires human review before any consequential action. It signs its requests with Web Bot Auth, with a Signature-Agent header of "https://chatgpt.com", and OpenAI does not document a user agent string for it.
Key takeaways
- ChatGPT agent acts for one person, so it is not a crawler.
- OpenAI says its cloud browser needs no allowlisting, but security tools may block it.
- Trust the signature, not the user agent string.
What is ChatGPT agent, and what does it do on a website?
ChatGPT agent works through a cloud browser. OpenAI's guidance for site owners says the browser "lets ChatGPT read pages, click buttons, and interact with your site on a user's behalf."
In practice, that means tasks such as finding products, comparing options and filling out forms. A person asks ChatGPT to do a job. The agent then visits sites the way a person would, one page at a time, with a real browser.
This is different from a crawler. A crawler collects pages in bulk. ChatGPT agent visits because one specific person asked it to get something done.
A note on names. OpenAI's pages for site owners use the term "cloud browser". Some security vendors list the same traffic as "ChatGPT Agent". How this relates to ChatGPT Atlas is not explained in the OpenAI pages we could read, so we do not describe it here.
Can ChatGPT agent buy things on my store?
Yes, where ChatGPT Wallet is available. OpenAI's cloud browser guidance does not describe checkout, but its merchant guide on accepting payments says: "Customers can use ChatGPT Wallet to pay on your website or use an existing saved payment method. The customer reviews and approves the purchase in ChatGPT, and your checkout processes the payment."
9to5Mac reported that ChatGPT Wallet launched on November 13, 2026, through OpenAI's dots agents, for ChatGPT Pro subscribers in the US aged 18 or older. OpenAI says your store receives a network token, not the full card number. For what to check, read ChatGPT Wallet: what merchants need to know.
What it does say is this: "Customers are ultimately responsible for the actions they take." If a payment or order goes wrong, OpenAI says to handle it as you would if the customer had acted directly.
So treat an order that came through the agent as a normal order from a customer. It is worth reading your own order data to see which orders arrived this way.
Does ChatGPT agent ask the user before it acts?
Yes, for consequential actions. OpenAI states: "across all our users, we require human review before any consequential action."
Two limits are worth knowing:
- Site owners currently cannot require human confirmation for specific actions on their own site.
- OpenAI's site-owner pages do not list which actions count as consequential. Do not assume a particular step, such as the final payment click, is always gated. The exception is ChatGPT Wallet: OpenAI says the customer reviews and approves each Wallet purchase in ChatGPT.
If you need a hard rule about what an agent may order, put it on your side. Order rules that approve, decline or hold an order for review work whatever the agent does.
How does ChatGPT agent identify itself?
It signs its requests. OpenAI says the cloud browser uses Web Bot Auth, built on the HTTP Message Signatures standard, RFC 9421. Each signed request carries three headers:
SignatureSignature-InputSignature-Agent
OpenAI says Signature-Agent must be set to "https://chatgpt.com", including the quotation marks.
Signature-Agent: "https://chatgpt.com"
Signature-Input: ...
Signature: ...
The public key lives at https://chatgpt.com/.well-known/http-message-signatures-directory. A site checks the signature against that key.
OpenAI's guidance does not mention a user agent string or IP address ranges for the cloud browser. It tells you to use signature verification instead. If you filter on a user agent alone, you cannot tell the real agent from a request that only claims to be it. For a deeper look at checking visitors, see how to tell if a visitor is really ChatGPT.
How is ChatGPT agent different from GPTBot, OAI-SearchBot and ChatGPT-User?
They have different jobs, and OpenAI controls them separately. OpenAI's crawler page describes them like this:
| Bot | What OpenAI says it is for |
|---|---|
| GPTBot | Crawls content that may be used to train OpenAI's generative AI foundation models. |
| OAI-SearchBot | Surfaces websites in search results in ChatGPT's search features. |
| ChatGPT-User | Used for user-initiated actions in ChatGPT and Custom GPTs. Not used for automatic crawling. OpenAI says robots.txt rules may not apply to it. |
| ChatGPT agent (cloud browser) | Signed browser requests that act on a user's behalf. Not described as a web crawler. |
OpenAI publishes IP lists for GPTBot, OAI-SearchBot and ChatGPT-User. The pages we read do not document an IP list for the cloud browser, and the crawler page does not mention ChatGPT agent at all.
OpenAI also says its crawlers, including OAI-SearchBot, are separate and have their own controls. Blocking GPTBot in robots.txt does not change what the cloud browser can do. For the crawler side, see robots.txt for AI agents.
Do I need to allow ChatGPT agent on my site?
Usually not. OpenAI says the cloud browser "can already use your site. You don't need to add it to an allowlist." Your own security settings may still block it.
If they do, OpenAI lists these steps by provider:
- Cloudflare: set "dots, OpenAI" to Allow under AI Crawler and Agent access policies. Or write a WAF custom rule using detection ID
129220581(bot tagchatgpt-agent). - Akamai: create a custom bot category with a Verified Bot named ChatGPT Agent and set it to Allow. Akamai verifies signatures automatically.
- HUMAN: turn on the ChatGPT Agent rule under Known Bots & Crawlers and set it to Allow.
- Vercel: no setup is needed unless your own rules block it. It is identified as
chatgpt-operator. - Other firewalls: verify requests by hand.
OpenAI's manual steps are:
- Check that the
Signature-Agentheader exactly matches"https://chatgpt.com". - Fetch the public key from the key directory above.
- Verify the
SignatureandSignature-Inputheaders under RFC 9421. - Only then apply an allow exception for signed browser requests.
An allow exception applies to the whole site. Keep your page-level access controls in place.
How do I tell the real agent from a fake?
Verify the signature. A request that only claims to be ChatGPT agent, with a matching user agent or a copied Signature-Agent header, proves nothing on its own. Only a signature that checks out against OpenAI's published key counts. Treat requests that claim to be ChatGPT agent but fail verification as unverified.
Cloudflare's documentation describes what a verifier checks: the Signature-Agent value is a quoted https:// URL, it is listed in the signed components, the signature has not expired, and the signature verifies against the registered key.
How Sill helps
Sill shows you which AI agents and bots visit your site and the pages they read. It verifies signed agent requests (Web Bot Auth) on Cloudflare and WordPress installs. When an AI agent places an order, your rules approve it, decline it, or hold it for a person to review, and every decision is signed and recorded. To see how ready your site is, run the free scan.
Questions people ask
Does ChatGPT agent obey robots.txt?
OpenAI's guidance on the cloud browser says it applies to signed browser requests, not web crawlers. OpenAI's crawler page says robots.txt rules may not apply to ChatGPT-User, which also acts on a user's behalf. OpenAI does not document robots.txt handling for the cloud browser.
Can I stop ChatGPT agent from using a page on my site?
OpenAI says to keep your existing page-level access controls, because an allowlist exception applies site-wide. Use your firewall or bot rules to block specific pages.
Can ChatGPT agent complete a checkout on my store?
It can pay where ChatGPT Wallet is available. OpenAI's merchant guide says customers can use ChatGPT Wallet to pay on your website after reviewing and approving the purchase in ChatGPT, and your checkout processes the payment. 9to5Mac reported that Wallet launched on November 13, 2026, for US ChatGPT Pro subscribers using dots. OpenAI also says customers are responsible for the actions they take.
Is ChatGPT agent the same as ChatGPT Atlas?
OpenAI's site-owner pages we could read do not explain how the two relate. We have left that out rather than guess.
Sources
- OpenAI: Help ensure cloud browser access to your site · checked 17 Nov 2026
- OpenAI for Developers: Accept payments · checked 17 Nov 2026
- OpenAI for Developers: Commerce partner playbook overview · checked 17 Nov 2026
- 9to5Mac: OpenAI launches ChatGPT Wallet, enabling 'dot' agents to make purchases · checked 17 Nov 2026
- OpenAI: Overview of OpenAI crawlers · checked 15 Nov 2026
- Cloudflare: Web Bot Auth · checked 15 Nov 2026