How to Use ChatGPT in Chrome: Let It Read Your Open Tabs, Work in the Side Panel, and Stay in Control
Image: OpenAI (official) — frame from the @ChatGPT launch video (Oct 15, 2026), showing ChatGPT comparing sofa listings across open Chrome tabs
On October 15, 2026, OpenAI pushed a bigger update to its ChatGPT browser extension than the version number would suggest. In OpenAI’s words, it is “bringing the power of computer use to anyone who uses Chrome.” Two things changed. First, ChatGPT can now use the tabs you already have open: in Chat it looks across them to gather information, and in Work it can operate them, sending messages, filling in forms, downloading files, and even grouping that pile of tabs. Second, the panel inside Chrome is now the full ChatGPT, with your whole conversation history, side by side with the page you are on.
This guide goes through setup, the three ways to give ChatGPT page context, how to run a real browser task, and the permission settings worth deciding on before you click “Allow for all sites.” Every step below is taken from OpenAI’s own documentation and announcement posts.
1. What you need first
- A ChatGPT account. OpenAI says the extension is available on all ChatGPT plans. The ability to act on pages goes through ChatGPT Work, which has its own plan availability, so check that you can see Work in your account.
- A supported browser. Chrome, Microsoft Edge, Brave, and Vivaldi get the full feature set, including side chat. Opera supports tab mentions and browser control from the desktop app, but no side chat.
- The ChatGPT desktop app, kept up to date (macOS, Windows, or the Linux preview). OpenAI’s setup flow starts there, and the troubleshooting guide’s first step is always “update the desktop app.”
- The right browser profile. The extension only works in the profile where you install it. If you split work and personal Chrome profiles, choose one on purpose.
2. Install and connect the extension
You can install the extension straight from the Chrome Web Store listing, which shows about 7,000,000 users. OpenAI’s documented route goes through the desktop app instead, so the connection is set up properly:
- Open the ChatGPT desktop app and go to Settings > Computer Use. If your browser isn’t in the main list, expand More browsers.
- Select your browser and follow any prompt to install the required plugin.
- Click Install next to the browser. That opens the extension store page. Add the extension and read the permission prompt (more on that below).
- Go back to Computer Use and check that the browser now shows Manage rather than Install. That is your sign that the connection works.
The browser’s toggle on that screen decides whether it appears in the @-mention menu. Manage is where you control which websites it can use.
3. Three ways to hand ChatGPT a page
Side chat (Chrome, Edge, Brave, Vivaldi). Open the page you care about, then click ChatGPT in the toolbar or the Extensions menu. On a Mac, Cmd+Shift+. also works. The panel stays attached to that tab. Chats you start there show up in the ChatGPT app, and you can open recent chats inside the panel, so you can move between the two without losing the thread.
Mention a tab. In the desktop app, and in side chat, you can mention an open tab so ChatGPT uses that page as context. This is the new multi-tab part. Open five pricing pages, mention them, and ask one question across all of them, with no copying and pasting.
Selected text or right-click. Highlight a passage and bring just that selection into the chat, or right-click the page and choose Ask ChatGPT. The side chat opens with that page’s context already loaded. On YouTube, side chat can use the video’s timestamped transcript when captions exist, which helps for “where in this talk do they cover X?”
4. Run a real browser task
Reading tabs is Chat territory. Doing things in them is Work territory. To start a task:
- Start a new ChatGPT Work (or Codex) chat in the desktop app.
- Type
@and choose Chrome (or Edge, Brave Browser, Opera, Vivaldi) so ChatGPT knows which browser to drive. - Describe the outcome and the point where it must stop.
OpenAI’s own example prompts make good templates, and the best ones say clearly where to stop:
- “Compare the three couches I have open by dimensions, material, delivery time, and return policy.”
- “Compare these pricing pages against our requirements. Which plan includes everything we need?”
- “Use the details in this event brief to fill out the registration form, and stop before submitting.”
- “Put the vendors in these tabs into my comparison sheet, with pricing, key features, and source links.”
Notice the “stop before submitting.” My advice is to put an explicit stop point in every action prompt for the first few weeks. The agent is good at filling fields. You should still be the one who presses the button that spends money, sends a message, or files something official.
5. Decide your permission level before you need it
By default, ChatGPT asks before it touches each new website, keyed on the host (for example example.com). When it asks, you get four choices: Allow once, Allow for this site, Allow for all sites, or Decline. OpenAI marks “Allow for all sites” as an elevated risk, because after that ChatGPT no longer asks before using any site open in the browser.
A sensible setup for most people:
- Use Allow for this site for the few tools you use every day, such as your expense app, CRM, or project tracker.
- Use Allow once for shopping, travel, and anything you visit rarely.
- Put banking, payroll, admin consoles, and your password manager’s web vault on the blocklist in Settings > Computer Use > Manage. The allowlist and blocklist are shared across all supported browsers.
- Treat browser history as a separate decision. ChatGPT asks each time it wants history, and there is no always-allow option. OpenAI warns that history can contain internal URLs and search terms, and that a malicious page could try to get them copied somewhere unintended.
About the install prompt: Chrome will list permissions like “read and change all your data on all websites,” “access the page debugger,” and “manage your downloads.” That sounds alarming, and it is broad, but it is what any extension that operates a browser needs. OpenAI says ChatGPT’s own confirmations, allowlists, and blocklists still apply on top. OpenAI also says it doesn’t keep a separate full log of your browser actions. Browsing data is stored only when it becomes part of the chat, such as page text it reads, screenshots, or tool calls.
6. When it won’t connect
OpenAI’s checklist, in order: make sure the site isn’t on your blocklist; update the desktop app (and remove duplicate ChatGPT or Codex apps); restart the browser and confirm side chat loads; confirm Manage shows under Computer Use and the toggle is on; check you are in the profile that has the extension; start a new Work chat; restart the desktop app; reinstall from Computer Use. If a task needs to upload a local file, open Chrome’s Manage Extensions, go to Details on the ChatGPT card, and turn on Allow access to file URLs.
My take
This is basically what the retired Atlas browser promised, without asking anyone to switch browsers. That matters more than any single feature. The Chat side (asking across open tabs) is low-risk and immediately useful for comparison shopping, vendor research, and reading long documents. The Work side is where you should go slowly. It acts inside your real, signed-in sessions, so a mistake is an actual sent message or submitted form, not a sandbox error. Start with read-only questions, give every action prompt a stop point, and keep your sensitive domains on the blocklist from day one.
Sources
- @ChatGPT on X: Chrome tabs announcement and demo video (Oct 15, 2026)
- @ChatGPT on X: full ChatGPT in the Chrome side panel (Oct 15, 2026)
- @ChatGPT on X: example prompts
- OpenAI docs: Browser extension (setup, permissions, troubleshooting)
- ChatGPT Training: Using your computer and browser
- OpenAI Help Center: ChatGPT release notes
- Chrome Web Store: ChatGPT extension
댓글
댓글 쓰기