ChatGPT Atlas Is Retired: What It Means for Site Owners and What to Do Instead
ChatGPT Atlas SEO is no longer about Atlas itself: OpenAI's help center now says ChatGPT Atlas has been retired and no longer receives updates. What still matters for site owners is everything Atlas relied on: OAI-SearchBot access for ChatGPT search, pages an AI agent can actually operate, and letting ChatGPT's newer cloud browser through your firewall if you want it. This guide separates what changed from what didn't, using only OpenAI's own documentation.
Short answer: Atlas is gone, the work it pointed to is not
OpenAI launched ChatGPT Atlas, a web browser with ChatGPT built in, on October 21, 2025. Today the original launch post carries a note that Atlas "has since been deprecated", and the help center's ChatGPT Atlas retirement article (updated October 23, 2026, when we checked) says it has been retired and no longer receives updates. Older Atlas help pages, including the publisher FAQ, now redirect there. OpenAI's pages we checked do not give a retirement date.
So if you were planning "Atlas SEO" work, there is no Atlas-specific ranking to chase. But the reasons SEOs paid attention to Atlas are still live:
- ChatGPT still searches the web and cites sources, and access for its search crawler still decides whether you can appear.
- ChatGPT still browses sites on users' behalf, now through a cloud browser in ChatGPT Work.
- Pages that an agent can read and operate (clear structure, labeled forms, real buttons) still have the edge OpenAI pointed to at Atlas's launch.
What Atlas was, and why site owners cared
From the launch post, three things mattered to website owners:
- Agent mode. ChatGPT could open tabs and click through sites to complete tasks such as filling a shopping cart. That made your site's usability for software, not just for people, a real factor.
- ARIA advice. OpenAI wrote that website owners could add ARIA tags to improve how ChatGPT agent worked on their sites in Atlas. It was one of the rare times an AI company gave site owners a concrete markup recommendation.
- Training and GPTBot. OpenAI said browsed content was not used for training by default, and that even when users opted in, pages that opted out of GPTBot would not be used for training.
Of those three, the browser is gone, but agent browsing and the GPTBot rule live on in other OpenAI products, covered below.
What did not change: ChatGPT search visibility
Atlas was one way to reach ChatGPT. The search side runs through ChatGPT itself, and its rules are unchanged. OpenAI's help article on searching the web with ChatGPT says that to make a site eligible you should allow OAI-SearchBot to crawl it and make sure your host or CDN allows traffic from OpenAI's published search bot IP addresses. It also says placement is not guaranteed.
The bots, per OpenAI's crawler documentation:
| Agent | What it does | How you control it |
|---|---|---|
OAI-SearchBot | Crawls sites for ChatGPT search | robots.txt; IP list at openai.com/searchbot.json |
GPTBot | Crawls content that may be used to train OpenAI's models | robots.txt; disallowing it signals that content should not be used for training |
ChatGPT-User | Fetches a page when a user's action in ChatGPT requires it | OpenAI says robots.txt rules may not apply, since a user started the request |
| ChatGPT Work cloud browser | Operates sites on a user's behalf in a remote browser | Signed requests you can verify and allow or block at your CDN (next section) |
If you only fix one thing, fix the first row. Run your domain through our AI crawler checker to see whether robots.txt allows OAI-SearchBot, and read how to get your website indexed by ChatGPT for the full process.
Where agent browsing lives now: ChatGPT Work's cloud browser
OpenAI's help page for ChatGPT agent now states that it is no longer available and points to ChatGPT Work. According to Using cloud browser in ChatGPT, the cloud browser gives ChatGPT Work its own browser on a separate computer in the cloud. It can read pages, click buttons, fill in forms and work on public and signed-in sites. It is available on paid plans in supported regions, excluding Free and Go.
The same article notes that some websites block automated browser agents and that each website decides whether to allow cloud browser traffic. In other words, this is now a setting on your side.
How the cloud browser identifies itself, from OpenAI's cloud browser allowlisting guide:
- It signs outbound requests with HTTP Message Signatures (RFC 9421), a scheme OpenAI calls Web Bot Auth.
- Each request carries
Signature,Signature-InputandSignature-Agent: "https://chatgpt.com"headers. - Public keys are published at
https://chatgpt.com/.well-known/http-message-signatures-directory.
OpenAI lists vendor-specific steps:
| Provider | How it appears | What to do |
|---|---|---|
| Cloudflare | Signed agent, bot tag chatgpt-agent, detection ID 129220581 | Create a WAF custom rule on that detection ID with Allow or Skip |
| Akamai | "ChatGPT Agent" in the AI bots category | Add it to a custom bot category set to Allow |
| HUMAN | "ChatGPT Agent" in Known Bots & Crawlers | Turn the rule on and set it to Allow |
| Vercel | Recognized as chatgpt-operator | No setup needed, per OpenAI |
| Other CDNs | Signature headers only | Verify the signature against the published key, then allow |
OpenAI's allowlisting page does not mention robots.txt for the cloud browser, and its crawler page does not list it. We would not assume robots.txt controls it. If you want to block it, do it at the CDN or firewall level, using the same identifiers.
Should you let AI agents operate your site?
It depends on what visitors do on your site. A simple way to decide:
| Your site | Sensible default | Why |
|---|---|---|
| Blog or content site | Allow | An agent reading your page for a user is a visit you would otherwise not get |
| Local business, bookings, quotes | Allow | OpenAI's own task examples include requesting quotes from businesses and checking restaurant availability |
| Online store | Allow, monitor checkout | OpenAI lists comparing product availability as a task; agents that cannot get in cannot recommend your stock |
| Paywalled content, member areas | Decide deliberately | The cloud browser can work on signed-in sites when the user signs in themselves |
| Site under scraping or abuse load | Allow verified, block unverified | Signatures let you tell the real agent from impostors using its name |
Make your pages usable by agents
OpenAI's only public markup advice for agents was the ARIA note at Atlas's launch. The W3C's own Using ARIA guide adds the key caveat: its first rule is to prefer a native HTML element when one already has the semantics and behavior you need. For a small site, that translates into:
- Real controls. Use
<button>and<a href>, not clickable<div>s. An agent, like a screen reader, sees what an element is from its markup. - Labeled forms. Every input has a visible
<label>. Error messages are text near the field, not just a red border. - ARIA only where HTML falls short. Custom menus, tabs and dialogs need roles and states (
aria-expanded,aria-controls), or an agent can't tell whether a menu is open. - Key facts as text. Prices, opening hours, stock and contact details should be in the HTML, not only inside images.
- Predictable flows. Avoid steps that only appear on hover, and don't require a CAPTCHA for simple read-only pages.
None of this is an OpenAI ranking factor; OpenAI hasn't published any. It is plain accessibility, which helps every visitor and every assistant that reads pages. For machine-readable facts, see our guide on schema markup for AI search.
Measuring ChatGPT visits after Atlas
OpenAI never documented a dedicated Atlas user-agent string in the pages we reviewed, so there is nothing Atlas-specific to filter for in old reports, and nothing to set up now. What you can measure:
- Search clicks. Visits from links in ChatGPT answers show up as referrals from chatgpt.com. Our guide to tracking AI traffic in GA4 shows how to group them.
- Crawler activity. Search server logs for
OAI-SearchBot,GPTBotandChatGPT-User, and confirm IPs against OpenAI's JSON lists. - Cloud browser visits. If your CDN logs request headers, look for
Signature-Agent: "https://chatgpt.com". Cloudflare and Akamai show these visits under the bot names in the table above. OpenAI doesn't document whether they execute your analytics script, so don't expect them to appear in GA4.
Checklist for site owners
- Remove "optimize for Atlas" tasks from your plan; replace them with the items below.
- robots.txt allows
OAI-SearchBot; GPTBot is allowed or blocked on purpose. - CDN and host allow OpenAI's published search bot IPs.
- Decide on ChatGPT Work's cloud browser and set the matching CDN rule (allow verified, or block).
- Buttons, links and forms use native HTML with labels; custom widgets have ARIA roles and states.
- Prices, hours and contact details are text in the HTML.
- Reports separate chatgpt.com referrals; logs are checked monthly for OpenAI bots.
Visibility in ChatGPT answers also depends on content and mentions elsewhere on the web, which we cover in how to get cited by ChatGPT and the AI search hub.
What to do next
The fastest check is free: enter your domain in the AI crawler checker and see, for OAI-SearchBot, GPTBot, ChatGPT-User and nine other AI bots, whether your robots.txt allows them and which line decides it. Then work through the CDN and markup items above, which robots.txt can't show.
FAQ
Is ChatGPT Atlas still available?
No. OpenAI's help center says ChatGPT Atlas has been retired and no longer receives updates, and the original launch post is marked as deprecated. The help center explains how to move bookmarks you exported earlier into another browser.
Do I need to optimize my site for ChatGPT Atlas?
No. With Atlas retired there is nothing Atlas-specific to optimize. The things that still matter are allowing OAI-SearchBot for ChatGPT search, deciding whether to allow ChatGPT Work's cloud browser, and keeping pages accessible to agents.
What replaced ChatGPT agent mode?
OpenAI's help center says ChatGPT agent is no longer available and points to ChatGPT Work, which uses a cloud browser to read pages, click buttons and fill in forms on a remote computer. It is available on paid plans in supported regions, excluding Free and Go.
How do I allow ChatGPT's cloud browser on my site?
It signs requests with HTTP Message Signatures and a Signature-Agent header set to https://chatgpt.com. OpenAI documents allow rules for Cloudflare (detection ID 129220581), Akamai and HUMAN, and says Vercel needs no setup. Other CDNs can verify the signature with OpenAI's published keys.
Does robots.txt block ChatGPT's cloud browser?
OpenAI's documentation doesn't say. Its allowlisting guide only describes CDN and firewall rules based on signed requests. If you want to block it, do it there, not in robots.txt.
Does ARIA markup help with ChatGPT?
At Atlas's launch OpenAI said ARIA tags could improve how its agent worked on a site. OpenAI has not published ranking factors. Native HTML elements with labels come first, and ARIA fills the gaps for custom widgets.
Read next
Check your site's Quality Score
Free, no sign-up: a 0–100 score from mostly open data. Then add your site to Satelo and get partner articles from sites of your level.
Enter a domain like example.com
Add my site