Watch: Google Uses Chrome to Supply Context to Gemini Chat
An analysis of Chrome's Page Content Agent, which builds a structured tree of page elements to provide Gemini with grounded context for reading and interaction.
Transcript
When you share a Chrome tab with Gemini, the browser doesn't just send raw code. Instead, it runs an internal system called the Page Content Agent. This agent walks through the page's layout and builds a highly structured map of everything on your screen, from headings and paragraphs to buttons, images, and table cells.
Each element is tagged with a unique node ID, its exact location on the screen, visual styling, and how a user can interact with it. This map is then translated into clean Markdown with embedded node references. When Gemini receives this structured text, it can read the page, cite specific parts using the node IDs, and even interact with the site, like clicking buttons or filling out forms, by referencing those same IDs.
This process also respects your current session. Because the browser grabs the page exactly as it is rendered on your screen, Gemini sees what you see. If you are logged into a website, the AI can see your account dashboard, draft content, or even your bank balance if you ask it to analyze that tab.
To protect your privacy, Chrome built safeguards directly into the pipeline. Password fields are completely redacted, paid content behind paywalls is flagged, and embedded content from other websites is stripped out before the data ever leaves your device. Advertisements are also identified and filtered.
Ultimately, instead of sending a messy wall of code, Chrome packages your active tab into a clean, interactive map. This allows the AI to navigate, comprehend, and help you manage the web pages you are looking at in real-time.