Meta has begun testing an “AI Mode” inside Facebook search in the United States that answers users’ questions by drawing on public content across its platforms, including public posts, groups and Reels. The company has indicated the feature may expand to other markets over time.
The change turns Facebook’s search box from a tool for finding pages and people into something closer to an AI assistant. Ask about a local restaurant, a travel destination or a product, and the system is designed to compose an answer from what people have posted publicly, alongside web results.
What Meta has said
According to Meta’s description of the feature, reported by TechRepublic and The Next Web, AI Mode can use publicly available posts, public groups, Reels and other open content to produce more detailed responses to queries. The company has not spelled out in detail how it weighs different sources or how much of any given post may be reflected in an answer, but it has confirmed that public content is part of the mix.
There is, so far, no evidence that Meta AI reproduces specific individual posts verbatim in its answers. The concern raised by privacy advocates is about aggregation rather than quotation: information that was always technically public becomes far easier to find and combine once a system can read across millions of posts at once.
Three concerns
The first is privacy. Over many years, large numbers of people have written in public groups about their daily lives, purchases, travel, local problems, health conditions, pets and family events. Scattered across thousands of groups, that information was hard to assemble. Summarized by a model, it is not.
The second is fraud. Someone with bad intentions who wants to know what people in a particular city are currently worried about, buying or discussing could in principle surface those patterns faster with an AI layer on top of public posts, and then tailor scams, fake advertisements or phishing attempts accordingly.
The third is accuracy. Not every Facebook post rests on reliable information. If a system treats a user’s personal opinion and a published study as equivalent inputs, unverified or misleading claims can end up in its answers. Other AI search products have already run into this problem when they drew on internet forums and returned strange or incorrect advice.
How to limit exposure
Users who want to reduce how much of their content is available to these systems have a few straightforward options:
- Change the audience on older public posts to “Friends” using Facebook’s privacy tools.
- Set the default audience for future posts to “Friends” rather than “Public”.
- Avoid sharing in public groups anything you would not want an AI system to read, summarize or learn from later.
None of these steps affects content that others have posted about you, and they do not retroactively remove information already gathered. But as search on social platforms becomes AI-driven, the line between “visible to anyone who looks” and “findable by anyone who asks” is disappearing, and privacy settings that once felt like a formality now carry more weight.
