New ChatGPT Feature Can See Which Websites Users Have Open

OpenAI has released an update that allows ChatGPT to read Google Chrome tabs that are open and help users complete tasks directly through the browser.

The feature means the AI chatbot can do more than answer questions: it can compare information and perform a range of tasks directly. The update offers two modes: Chat mode and Work mode.

In Chat mode, ChatGPT can read multiple open tabs and summarise their contents. Users can ask the AI to compare products across several websites, including by size, material, delivery time and return policy.

ChatGPT can also simplify the content of articles or compare subscription plans without requiring users to switch between browser windows.

Work mode, meanwhile, allows the AI to take direct action to complete tasks. Users can ask ChatGPT to fill in online registration forms, draft emails, send messages, download files and organise browser tabs.

The AI can also gather information from different websites and arrange it in a simple comparison table, India Today reported on Tuesday (20 October 2026).

To use the feature, users need to install the ChatGPT desktop app and the official ChatGPT extension from the Chrome Web Store. They must then enable the Computer Use feature in the ChatGPT app settings.

Users can then start a new conversation in Work mode, type the “@” symbol to select a browser, and give ChatGPT an instruction. For example, they can ask the AI to compare mobile phone prices across several open tabs.

The feature can also be used with other browsers, including Microsoft Edge, Brave and Opera.

Although the feature offers convenience, giving AI access to a browser also creates security risks. Since ChatGPT can read information displayed on screen, users are advised to close sensitive tabs, such as online banking or private work email, while the AI is controlling the browser.

OpenAI has added safeguards requiring user permission before the AI interacts with a new website or downloads a file. However, the risk of prompt-injection attacks remains.