Apple tests Siri web answers, may use Google’s Gemini

October 3, 2025

ahmad_superadmin_user

Apple is developing an AI-powered search feature for Siri that could rely on Google technology, according to a report from Bloomberg’s Mark Gurman. Internally called “World Knowledge Answers,” the tool would let Siri look up information on the web and return short, readable summaries. Apple is said to be considering a custom version of Google’s Gemini model, hosted on Apple’s servers, to generate those answers.  The interface would combine text with photos, videos, and points of interest so results feel richer and more useful, giving Siri a way to compete with AI search experiences from OpenAI and Perplexity. Apple’s work on AI search sits within a larger Siri upgrade that has been delayed but remains a priority.  The new system is expected to use three main parts: a planner that interprets voice or text prompts, a search layer that can scan personal data or the web, and a summarizer that packages the response in clear language. Bloomberg reports that Apple and Google reached a formal agreement this week that allows Apple to test a Google-designed model for Siri’s summaries.  For searches across user data, Apple still plans to rely on its own models. The company is also evaluating Anthropic’s Claude and Google’s Gemini for Siri’s planning function, indicating that Apple may mix providers for different tasks while keeping tight control over privacy and performance. While Apple is expected to unveil the iPhone 17 lineup next week, the AI-enhanced Siri is not slated to ship immediately with that hardware.  Instead, the upgraded assistant could arrive alongside iOS 26.4 as early as March, giving Apple time to test the new flow and scale server capacity. If released as described, World Knowledge Answers would turn Siri into a more helpful everyday tool by pulling trusted web results into quick summaries and by understanding on-screen context to perform actions.  It would also mark a notable shift for Apple: a willingness to lean on outside AI models for certain capabilities while building its own for tasks that touch personal data.