Full-screen AI with its own browser, use your own Claude plan, and updates to /goto links

Full-screen AI with its own browser

AI now has its own full-screen workspace, alongside the side panel. Open it from AI in the left rail (Alt+3), Go to AI in the command palette, or Open full screen in the AI panel. Everything the AI panel does is available here, and both views share one chat history. The chat list sits beside the conversation and shows which chats are running, need you, have failed or have stopped. Up to 3 chats can keep running while you work in another, and archived chats have their own section at the bottom of the list.

The AI now opens pages in its own browser, so it never takes over your tabs. Those pages use your browser profile, which means it's signed in wherever you are. Select 'Show AI browser' through the globe icon in full screen to see the page it's working on or enter an address yourself. Tabs you attach to a chat still give it context.

Use your Claude plan in the AI panel

Following interest from users, you can now run Claude models on your own Claude plan, just as you can with GPT models and a ChatGPT plan. Connect in Settings → AI, or from Billing → Pay with in the model picker. The first connection downloads Claude Code to your device (about 290 MB), then you sign in through your browser or with a code.

Models running on a connected plan don't use AI credits, which makes for more efficient usage. The Pay with menu lists AI credits once, with an on/off row for each plan, so you can use your Claude plan for Claude models and leave GPT models on credits. Plan limits and ended sign-ins are shown in the interface.

Attach and view files in AI chat

You can now attach files to a message in AI chat, so the AI can read and work with them alongside your pages, search results and connected data.

A new file viewer also lets you open CSV, Markdown, PDF and image files directly in SERP Lens.

Changes to Google's /goto links

On 30 October, Google began limiting how quickly its /goto links can be resolved, which stopped the workarounds many SEO tools used to reveal destination URLs, including ours. On each results page, the first 6–7 lookups return the real URL instantly. After that, Google shows a "Verifying your request" page, and the wait grows with each lookup.

SERP Lens now works with this delay. 'Who owns page 1' lookups and top organic results resolve first; the remaining links wait out Google's delay and retry, and other links resolve when you hover over them. Queries with around 6 organic listings have no delay. On busier results, the last links can take from several seconds to a minute or more, and the remaining domains fill in without re-running anything. 'SERP Feature' counts aren't affected. An AI agent reading a page as it loads can still receive /goto URLs for links that haven't resolved yet.

Sign in to skip the wait: you can sign in to your Google account in the SERP Lens browser. In our testing, every result link then arrived directly, with no /goto links to resolve. Google normally personalises results for signed-in accounts, and searches, including those the AI runs, are saved to that account's Google search history. We're working on keeping signed-in results unpersonalised by default.

Rank tracking: we've replaced the /goto links stored in rank tracking history with their real URLs and recalculated positions. Keywords and competitors that were wrongly showing as not ranked now show their correct position. New results are resolved as they're collected, though some can still fail. We're also actively working on the same kind of clean-up for poisoned SERPs.

Real SERP impact: there has been some talk surrounding poisoned or fake SERPs impacting real search results for users more broadly, contributing to ranking drops that could impact traffic for websites. Based on our own live SERP data analysis, this goes against what we have found at SERP Lens. According to the results of our analysis, the results were very similar across the board, with there being a very low likelihood of SERP poisoning or results looking real on the surface but being completely irrelevant.

Google has changed this repeatedly since June, so we'll keep testing and share updates as things change.

Also in this release

• AI search by country: the AI can now search by country and language without a VPN.
• MCP server: version 1.4.0 follows shortly after this release, adding tools for AI search answers, device size and user agent, and taking the total to 35.
• Long chats on a ChatGPT plan: chats no longer stop with an error once they reach the point where their history has to be shortened.
• MCP connections: connections that use sign-in now stay signed in after a token refresh.
• AI top-ups: a top-up now only charges the team it started on.
• Privacy: page URLs now stay out of crash reports and analytics.
• Links: link data now follows a page's <base href>.
• Shortcuts: AI is now Alt+3 in the left rail, and Shots moves to Alt+4.
• Polish: we've made a broad visual pass across the AI workspace and panel, including tables, the composer and dark mode.