Google has begun rolling out a new feature designed to improve the transparency of its AI-powered search results, allowing users to quickly verify the sources used to generate summaries in both AI Overviews and AI Mode. The update, announced Thursday by Google Search product VP Robby Stein on X, displays links in pop-up windows when hovered over, providing a description and direct access to the original websites.
The change addresses a key concern regarding the employ of artificial intelligence in search: the potential for inaccuracies and the difficulty in tracing information back to its origins. Previously, Google displayed a list of sources on the right side of the screen, but the new pop-up windows offer a more granular connection between specific claims within the AI summary and their corresponding sources, according to Stein.
“Our testing shows this new UI is more engaging, making it easier to get to great content across the web,” Stein wrote. The feature is now live for users accessing Google’s search engine via desktop browsers.
The rollout comes as Google continues to refine its Gemini-based AI offerings. Gemini 3.1 Pro, the latest iteration of the model, was released in preview earlier this month, boasting improved reasoning capabilities and enhanced performance on benchmarks like ARC-AGI-2, where it achieved a score of 77.1%, more than doubling the previous result. According to ZDNET, the new model builds on the progress established by Gemini 3, which outperformed competitor models like Microsoft’s Copilot in some in-house tests.
Google’s move towards greater source transparency also coincides with the broader release of Gemini 3, described by Google CEO Sundar Pichai as the company’s “most intelligent model.” Gemini 3 is now available across Google products, including the Gemini app, AI Studio, and Vertex AI. A more specialized version, Gemini 3 Deep Think, is slated for release to Ultra subscribers, focusing on complex research challenges in fields like chemistry and physics.
Although Gemini 3.1 Pro demonstrates gains in reasoning, it does not consistently outperform all competitors. Interesting Engineering reported that Claude Opus 4.6 currently leads Gemini in text-based tasks on the Arena benchmark, and several other models, including OpenAI’s GPT 5.2 High, rank ahead in coding categories. Arena rankings are based on user voting, which can prioritize outputs that *appear* correct, even if they contain subtle errors.
The update to Google Search’s AI features arrives after the release of Gemini 3 in November 2025, and follows a “major upgrade” to Gemini 3 Deep Think last week, according to Google. The company has not announced a firm date for the full rollout of the source-linking feature beyond the current desktop preview.
