ChatGPT’s updated Chrome extension lets users ask questions about their open tabs and delegate browser tasks, expanding the assistant from a conversational sidebar into a tool that can work with web pages. The update includes a Chat mode for gathering information across tabs and a Work mode that uses Computer Use to carry out actions in the browser.
Users can reference open tabs with @chrome, giving the assistant access to the pages relevant to a request. Chat mode can draw information from several tabs at once, supporting tasks that would otherwise require switching between pages and manually collecting details. The updated extension is available across all ChatGPT plans.
The distinction between the two modes is central to the update. Chat mode focuses on reading and synthesizing information, while Work mode can control tabs to complete tasks. Supported actions include sending messages, filling out forms, downloading files and organizing tabs into groups. These capabilities extend beyond generating an answer: they allow the assistant to change the state of a browser session or interact with a website.
OpenAI’s examples include comparing products across open tabs using criteria such as dimensions, materials, delivery times and return policies. Other examples involve summarizing articles in simpler language and comparing pricing pages from different sellers. Each relies on the browser as a shared workspace, rather than requiring users to copy individual passages into a separate conversation.
The upgraded sidebar also places users’ full ChatGPT conversation history alongside the page they are viewing. That brings previous conversations into the same interface as the current browsing session, while the tab-access features provide a way to introduce web content into a request.
Technically, cross-tab reading and browser control address different parts of a workflow. Comparing products requires extracting relevant information, aligning comparable attributes and identifying differences. Completing a form or sending a message adds an execution step, in which the assistant must interact with the page rather than simply describe what the user should do. Combining those functions can connect research and follow-through within one browser session.
Browser automation also introduces considerations that do not arise in a purely conversational exchange. Web pages may contain incomplete, outdated or conflicting information, and an incorrect interpretation can affect subsequent actions. Sending a message or submitting a form can have consequences outside the chat, making the distinction between reviewing information and authorizing an action important.
The supplied announcement details do not establish how the extension handles approval prompts, access restrictions or sensitive information. Those implementation details remain relevant for users evaluating which browsing tasks to delegate, particularly when open tabs contain private correspondence, account information or work documents.
Sources: The Decoder