Meta is turning years of social chatter into a new kind of search engine, promising quick, conversational answers built from Facebook and Instagram posts while raising fresh questions about accuracy, privacy, and power.

Early rollout and product design

On June 15, Meta began rolling out “AI Mode,” a new option in Facebook’s search bar that uses its Meta AI assistant to answer questions in plain language instead of showing a list of links.¹ The system pulls from public posts across Facebook, including Groups, Reels, and Marketplace listings, effectively turning “years of user-generated content into a searchable knowledge base.”¹

The feature is powered by Meta’s Muse Spark model and appears alongside existing filters like People and Marketplace.² Meta says it “uses Meta AI to give you answers grounded in what people are saying publicly across our apps” and plans to “over time unlock new features that cite recommendations and content people share across Instagram, Facebook, and Threads.”²

Part of a broader AI expansion

AI Mode arrives amid a broader push to embed AI across Meta’s products. The company has recently shipped AI-powered animated profile pictures, Marketplace auto-replies to buyer inquiries, and a creator assistant for captions and engagement.¹³ It also launched Forum, a Reddit-style app with an AI “Ask” tab that similarly summarizes discussions from Facebook Groups.¹³

Analysts note the timing parallels Google’s AI search overhaul, which has fueled “zero-click” behavior; Meta is now applying “the same approach to social content, synthesising public posts into AI-generated answers instead of sending users to the original discussions.”¹

Reliability, privacy, and user concerns

Reporters who tested AI Mode found it promising for local discovery but riddled with errors. In one hands-on, the tool mixed useful recommendations with “dodgy” justifications for January 6 rioters and incorrect details about nearby amenities, sometimes even suggesting activities in different states.⁴ The underlying risk, critics say, is that answers are distilled from “everyday users rather than vetted sources,” making outdated or misleading information more likely.³

Meta has not clarified whether group admins or users can opt their public posts out, or how it treats posts later made private or deleted — “significant gaps for a feature that treats user content as raw material for an AI system.”¹

As the rollout continues in the US, the central tension remains: can Meta turn the chaos of social feeds into trustworthy search without deepening concerns over misinformation and data exploitation?