ChatGPT Browser: the built-in browser vs the cloud browser
The desktop app's built-in browser and ChatGPT Work's cloud browser are two different things. This covers what each can do, the website permission model, annotation feedback, and why page content must be treated as untrusted.
Platforms
- ChatGPT desktop app (built-in browser)
- ChatGPT web and mobile (ChatGPT Work cloud browser)
What the official documentation says
Browser is available in ChatGPT on the web and in the ChatGPT desktop app, and is not available in Codex CLI or the Codex IDE extension.
BrowserThe built-in browser uses a browser profile separate from your regular browser and does not automatically share your existing tabs or browser session.
BrowserThe built-in browser opens from the toolbar, by clicking a URL, by navigating manually, or with Cmd+Shift+B (Ctrl+Shift+B on Windows).
BrowserThe documentation instructs you to treat page content as untrusted context and to review the site and the proposed action before sharing sensitive information or allowing ChatGPT to act.
BrowserChatGPT asks before using a website unless you have already allowed that site; allowed and blocked sites are managed in Settings > Browser.
BrowserChatGPT asks for confirmation before sensitive actions such as submitting information, making a purchase, changing permissions, or deleting data.
BrowserChatGPT cannot automate file uploads in the built-in browser.
BrowserThe built-in browser can be referenced directly with @Browser.
BrowserDeveloper mode gives ChatGPT controlled access to the Chrome DevTools Protocol and is enabled under Settings > Browser > Developer mode > Enable full CDP access; it cannot be enabled locally if the organization has disabled the setting.
BrowserChatGPT Work uses its own browser running on a separate computer in the cloud, not the browser on your phone or laptop.
BrowserThe cloud browser maintains its own cookies, browser data and signed-in sessions, and does not use your personal browser's open tabs, history, saved passwords, cookies, extensions or existing signed-in sessions.
BrowserCloud browser website permissions offer Always ask, Auto approve and Always allow, and the documentation explicitly does not recommend Always allow.
BrowserCredentials entered through the secure sign-in form go directly to the browser, are not visible to the model, are never used in model training, and ChatGPT does not store usernames or passwords.
BrowserCloud browsing is available in all regions on paid plans other than Free and Go, Enterprise admins must enable it for their workspace, and website sign-in is not available for Enterprise or Edu workspaces.
Browser
Two things are called "browser". Keep them apart
The most common source of confusion about ChatGPT and the web is that the documentation uses "browser" for two different products, with different capabilities, different privacy models and different jobs.
The built-in browser in the desktop app gives you and ChatGPT a shared view of websites and local web apps inside a chat. It is for previewing a page, leaving visual feedback, or letting ChatGPT interact with a site on your behalf. It runs on your machine and uses a profile separate from your regular browser — it does not automatically share your existing tabs or browser session.
ChatGPT Work's browser runs on a separate computer in the cloud, not the browser on your phone or laptop. It has its own cookies, browser data and signed-in sessions, and does not use your personal browser's tabs, history, saved passwords, cookies, extensions or existing signed-in sessions. Its job is to complete a task on real websites — and it keeps going after you step away and close your computer.
Decide which one you want before anything else; the configuration only makes sense afterwards.
One security premise you do not get to skip
The documentation states the requirement up front, and it is worth carrying verbatim: treat page content as untrusted context. Review the site and the proposed action before sharing sensitive information or allowing ChatGPT to act.
There is a sharper follow-up in the same page: a website permission lets ChatGPT interact with that site; it does not make the site's content trustworthy or approve every action.
That sentence exists because of a real cognitive trap — people grant a site permission and then treat everything downstream as safe. Instructions on a page can be misleading or malicious. Permission and trust are separate things.
The built-in browser: preview, annotate, act
Four ways to open it: the toolbar, clicking a URL, navigating manually, or Cmd+Shift+B (Ctrl+Shift+B on Windows).
Previewing a local page follows a standard loop: start your dev server in the integrated terminal or with a local environment action, open the route in the browser, review the rendered state alongside the code diff, leave browser comments on the elements that need changes, and ask ChatGPT to address the comments while keeping the scope narrow.
Annotation is the most useful capability here. Turn on Annotation mode, click an element or drag to select an area, write and save your comment, then ask ChatGPT to address it. The example comments in the documentation show the right register — not "this looks off", but the problem and the wanted result: "This button overflows on mobile. Keep the label on one line if it fits, otherwise wrap it without changing the card height."
There is a second layer for styling: select Adjust next to the text input on an annotation to change values such as font, text, spacing and colour, preview the result on the page, and send an annotation with a clearer target.
The boundaries on acting are equally explicit. ChatGPT asks before it uses a website unless you have already allowed that site, asks for confirmation before sensitive actions such as submitting information, making a purchase, changing permissions or deleting data, and cannot automate file uploads in the built-in browser.
Want it in your own Chrome? Different tool
If ChatGPT needs to work in an existing Chrome, Edge, Brave, Opera or Vivaldi tab, or use your regular browser profile, the documented answer is the browser extension, not the built-in browser. Developer mode works with both: @Browser for the built-in browser, @Chrome for Chrome.
The cloud browser: how sign-in actually works
This is the part of Work's cloud browsing worth getting exactly right.
When the browser hits a site that requires signing in, ChatGPT pauses and asks you to enter your credentials and any two-factor codes. The crucial fact: ChatGPT cannot see your username or password. They are never seen by the model or used in model training, and ChatGPT does not store them. Credentials entered through the secure sign-in form go directly to the browser.
There is a layer of protection above that too. Before ChatGPT asks you to sign in to any website, an additional review model checks the sign-in request and where your information will be entered for signs of phishing or deception. You see the website's address and a preview of its sign-in form, and you can inspect the live website before continuing.
The user-side rule is a single sentence: use the sign-in form ChatGPT provides, and do not send passwords in the chat.
Permissions come in three levels, set under Settings > Cloud browser: Always ask (review every request manually), Auto approve (ChatGPT approves after checking the website's relevance to your task), and Always allow (no additional review step). The third exists for minimal friction and the documentation says outright that it is not recommended. You can also allow or block individual websites to override the default.
To sign out, use Settings > Cloud browser > Browser data > Clear all. That signs you out of websites in Work's browser, so future tasks will need a fresh sign-in.
When to reach for something else
One practical steer from the documentation: for web apps you are building locally, use the built-in browser first. And when the task depends on a graphical interface rather than a web page, that is Computer Use territory.
How to do it
- Decide which browser you need. Use the desktop app's built-in browser to preview a locally running app and leave visual feedback; use ChatGPT Work's cloud browser to have ChatGPT complete a task on real websites, including ones that need a sign-in.
- Built-in — open it from the toolbar, by clicking a URL, by navigating manually, or with Cmd+Shift+B / Ctrl+Shift+B.
- Ask ChatGPT to use the built-in browser in your task, or reference it directly with @Browser.
- For precise feedback, turn on Annotation mode, click an element or drag to select an area, write and save your comment, then ask ChatGPT in the chat to address the comments.
- Manage allowed and blocked sites, browsing history and the download location in Settings > Browser.
- Cloud — describe the task in ChatGPT Work, approve website access when prompted, and when a login screen appears use the secure sign-in form rather than sending passwords in the chat.
- Set the website permission level and clear browser data in Settings > Cloud browser.
On Windows
- The keyboard shortcut for opening the built-in browser is Ctrl+Shift+B on Windows.
- Everything else — the separate browser profile, Settings > Browser for permissions and download location, and the @Browser reference — behaves as on macOS.
On mobile
- What is available on mobile is ChatGPT Work's cloud browser, not the desktop app's built-in browser.
- The documentation states you can start a task in Work on web or mobile and ChatGPT can keep working even if you step away and close your computer.
- On iOS you can use a supported password manager to sign in through the secure sign-in form.
Use cases
- Starting a local dev server, then asking ChatGPT to open a route, reproduce a layout bug and fix only the overflowing controls.
- Marking a UI problem on the page with a comment and having ChatGPT address it without touching the rest.
- Having ChatGPT Work complete a multi-step task on a site that requires signing in, such as comparing plans on your account.
Common mistakes
- Assuming the built-in browser carries your everyday browser's logins. It uses a separate profile and does not automatically share tabs or sessions.
- Sending a password in the chat. The documentation is explicit — use the sign-in form ChatGPT provides.
- Treating an allowed site as a trusted site. The documentation calls this out directly: a website permission lets ChatGPT interact with that site, it does not make the site's content trustworthy or approve every action.
- Expecting ChatGPT to upload a file for you in the built-in browser. The documentation states it cannot.
- Setting cloud browser permissions to Always allow. The option exists; the documentation says it is not recommended.
FAQ
- Are the built-in browser and Work's browser the same thing?
- No. The built-in browser lives in the ChatGPT desktop app and gives you and ChatGPT a shared view of websites and local web apps inside a chat. ChatGPT Work's browser runs on a separate computer in the cloud — not the browser on your phone or laptop — and maintains its own cookies and signed-in sessions.
- Can ChatGPT use sites I am already logged into?
- Not automatically, in either case. The built-in browser uses its own profile and does not share your existing tabs or session, though you can sign in directly inside it when a task needs an account. The cloud browser likewise does not use your personal browser's tabs, history, saved passwords, cookies, extensions or signed-in sessions; it keeps its own session, which stays active for future tasks once you sign in.
- What if I want ChatGPT to work in my normal Chrome?
- The documented answer is to use the browser extension instead. When ChatGPT needs to work in an existing Chrome, Edge, Brave, Opera or Vivaldi tab, or use your regular browser profile, that is the tool for it.
- What is developer mode for?
- It gives ChatGPT controlled access to the Chrome DevTools Protocol — profiling JavaScript, inspecting console output and network traffic, examining the DOM and applied styles, or diagnosing an issue in the live browser. Enable it under Settings > Browser > Developer mode > Enable full CDP access. Full CDP access can expose sensitive browser internals, so ChatGPT asks for explicit approval before using it on a website, and you cannot enable it locally if your organization has disabled the setting.
- Is cloud browsing available on my plan?
- The documentation states cloud browsing is available in all regions on paid plans other than Free and Go, that Enterprise admins must enable it for their workspace, and that website sign-in is not available for Enterprise or Edu workspaces. During rollout the browser might not appear immediately even when your plan supports it.
Official sources
These are the pages this tutorial is checked against. Follow them if you need the vendor's exact wording.
- Browser
https://learn.chatgpt.com/docs/browser.md