ChatGPT in Chrome: an assistant you can delegate browser work to

From "explain this page" to "do this in the browser"

Imagine: you have opened several services, and instead of manually transferring data, you ask an assistant to compile a comparison, check a form, or prepare an entry in a work system. This is exactly the kind of scenario OpenAI is developing an official extension for ChatGPT.

via the relevant Chrome Web Store card, it passes tab context and allows ChatGPT to perform actions in the browser. Stated scenarios include website research, working with documents, data entry, checking web pages, and organizing tabs. These are product capabilities, not a guarantee of executing any request.

Where to download and how to install

Start with the official ChatGPT download page: it has links to both the extension and the desktop app. Or open ChatGPT by OpenAI in the Chrome Web Store →. The name of a similar third-party app does not yet confirm its origin.

1. Install the extension in Chrome on your computer. Check the publisher is OpenAI, click the add button, and read the permissions.
2. Update ChatGPT desktop. Open Settings → Computer Use, select Chrome and follow the plugin installation prompts.
3. Check the connection. After Install, an icon should appear next to your browser Manage. Use the Chrome profile where the extension is installed.
4. Give your first task. In Work or Codex, select Chrome via the menu @. To ask about a page, you can open ChatGPT from the extensions panel next to it.

This order is described in the official Browser extension instructions. Having the extension does not mean unlimited access to all features: check the available capabilities in your account. The instructions refer to a computer, not installing a Chrome extension on a phone.

Six tasks to get you started

Below are our example prompts. They show how to set a specific task; this is not a report on six conducted tests. Start by reading an open page, then move on to changes.

1. Deconstruct a complex page

@Chrome Explain this page in Ukrainian for a beginner. Highlight 5 main ideas, unclear terms, and things the author is only assuming. Add a link to the source.

2. Compare tools without advertising clutter

@Chrome Compare these three open services: price, plan limits, export, and commercial use. Provide a table with links and the verification date. If a condition is not specified, write that as well.

It is useful to check not only the attractive price on the homepage, but also the pricing page, documentation, and the terms for the specific feature you need. This makes it easier to notice if an important export is only available in a different plan.

3. Check your website after Vibe Coding

@Chrome Check my website as a new visitor: menu, main links, and form clarity. For each issue, provide the page and reproduction steps. Do not submit the form, buy anything, or make changes.

Such a review complements the AI code check. Navigating through pages does not replace a security audit, server testing, or speed measurement.

4. Find materials for a 3D project

@Chrome Find 5 motion reference sets for running animation. For each, specify the author, angles, price, and terms of use. Distinguish between educational viewing and permission to include the material in a commercial product.

For an animator, this is a way to prepare a list for personal review. Assessing weight, rhythm, and movement character should still be done manually; a found description does not confirm the quality of the material.

5. Deconstruct a video tutorial

Summarize this YouTube tutorial in Ukrainian. Provide timecodes for key steps and a list of actions to repeat. Separately note anything that cannot be understood from the available subtitles.

OpenAI describes responses based on a transcript with timestamps, when subtitles are available. This is not a promise to see every cursor movement or evaluate the entire animation in the video.

6. Prepare data for a workflow system

@Chrome Based on these notes, prepare a CMS entry: title, short description, and tags. Show a draft before saving. Perform the publication only after my confirmation.

It is convenient to set a boundary immediately: prepare, save a draft, or publish. For your first time, choose a test entry that is easy to verify.

What's new and which browsers it works in

OpenAI Update Log As of August 25, 2026, it already describes support for Chrome, Edge, Brave, Opera, and Vivaldi. All five support tab mentioning and desktop control; Opera does not have a side chat. Described features also include passing selected text, the 'Ask ChatGPT' command via right-click, and YouTube questions.

The store card update from October 16 is fresher than old instructions that only described Chrome and a mandatory desktop scenario. At the same time, do not assume that every feature is simultaneously available to everyone: OpenAI explicitly notes dependence on rollout and workspace policies.

What permissions to grant

In ChatGPT settings there is control over access to sites and Browse, Download, and Upload actions. You can require confirmation or block access. An organization can set its own restrictions.

For the first attempt, our recommendation is to allow only the necessary site and give the task without changes. Check the active account before working with email or an admin panel. Enter passwords into the login form, not into the task text. For an important result, ask to show exactly what was done, and verify the final state yourself.

As explained by browser documentation, @Browser marks the built-in browser, while @Chrome — the connected Chrome. This matters when you need an already open session of a specific account. There is no need to turn on developer mode with full CDP for a standard first task.

If ChatGPT cannot see Chrome

Update the app, restart Chrome, check the profile and Manage status in Computer Use. If the browser is missing from the @ menu, check its toggle. Then try a new Work/Codex chat. Detailed steps are in the Troubleshooting section official instructions.

In OpenAI training material there is also a useful principle: if a task already supports a specific plugin, ChatGPT can use it; a browser is needed when the page interface is important. Therefore, task execution does not always look like a long series of clicks.