Facebook’s new AI Mode search gets its info from public posts

Meta is rolling out an AI Mode search on Facebook that generates answers from public posts across its platforms rather than returning traditional links, the company announced [1]. The feature appears alongside standard search filters such as "People" and "Marketplace" and is powered by Meta's Muse Spark AI model [1]. When a user submits a query, the system synthesizes information from publicly shared content on Facebook, Instagram, and Threads, then presents an AI-generated summary [1]. Users can pose follow-up questions directly within the search interface to refine results [1]. Meta stated the model will "over time unlock new features that cite recommendations and content people share across Instagram, Facebook, and Threads" [1]. The approach mirrors moves by competitors. Google Search, which commands roughly 90 percent of the global search engine market, now prefixes many results with "AI Overviews" that summarize information drawn from across the web [2]. Google has also surfaced content from Reddit threads within its AI-generated summaries [1]. Meta's deployment brings similar generative capabilities inside its own ecosystem, where it can draw on the activity of Facebook's approximately 3.07 billion monthly active users as of December 2023 [3]. Meta Platforms owns and operates Facebook, Instagram, WhatsApp, Messenger, and Threads, and derived 97.8 percent of its total revenue from advertising as of 2023 [10]. The company has invested heavily in artificial intelligence research, including the recent release of Meta SecAlign, an open-source large language model designed to resist prompt injection attacks [6]. The concentration of AI development within a small group of large technology firms has drawn scrutiny from antitrust regulators in the United States and Europe [7]. The AI Mode launch arrives alongside other AI-driven features, including photo presets that overlay sports jerseys onto fans and automated collage template suggestions [1]. Meta has not specified whether private posts or direct messages will ever inform the search results. The company's Messenger service, which counts over 1 billion monthly users, supports end-to-end encryption for conversations, a setting that would exclude those exchanges from any public-indexing system [11].