The new Facebook search feature, called "AI Mode," uses Meta AI to synthesize a single answer from public posts, Groups, and Reels, and asks the reader to trust a stranger's post the way they used to trust a link.
Meta has begun rolling out a new kind of search on Facebook. Instead of returning a list of links, the new "AI Mode" asks Meta's AI to read public posts, group discussions, and Reels, then write a single paragraph that it says answers the user's question. The feature was reported by TechCrunch on Monday and is the clearest signal yet that Meta is willing to put everyday user chatter in the place where search results used to be.
The mechanics matter more than the marketing. A user types a plain-language question into the Facebook search bar and gets a written answer back, grounded in public posts the AI picked from across Facebook, including Groups and Reels, rather than in vetted reference material or journalism. The product is, in TechCrunch's words, "a search experience that uses Meta AI to synthesize answers from public posts across Facebook, including Groups and Reels."
That is a different category of product from a link list, and it raises a different category of question. The substrate is no longer a page a journalist or a reference editor decided was worth your time. It is a stranger's public Facebook post, weighted by whatever signal Meta's systems use to pick it. The reader has to evaluate truth from a paragraph whose underlying sources are mostly invisible.
Three questions are worth carrying into any encounter with an answer like this. First, what is the substrate: are the sentences drawn from vetted reporting, reference work, and primary documents, or from a stream of public social posts, with all the staleness, rumor, and argument that implies? Second, is the synthesis transparent: can the reader see which posts the answer was built from, or does the paragraph arrive as a finished block with no visible seams? Third, what is the recourse if the answer is wrong or stale: can the user click through, challenge the underlying posts, or report a synthesized answer that misses the point?
Those are not hypothetical worries. The same experiment is already running on Google's AI summaries of Reddit threads, and the same reliability questions have surfaced there: outdated information, confidently phrased summaries of arguments that were never true to begin with, and no obvious way to see what the model actually read. Meta's version, if it lands the way the company hopes, scales that experiment across a much larger and more varied pool of public posts.
Meta is also pushing the same model in a second place. The company quietly launched "Forum" in May, a Reddit-style standalone app whose AI "Ask" tab pulls answers from Facebook Groups discussions. Two surfaces, one architecture: ask a question, get a paragraph, the paragraph is built from public social chatter.
The company frames both products as part of an effort to catch up with rivals in the consumer AI market and as a way to drive more engagement with its AI assistant. The watch item is whether the product comes with any visible scaffolding for the questions above: a list of underlying posts beside the synthesized paragraph, a way to flag a wrong answer, any accuracy disclosure. None of that appears in the current product description, and without it the reader is being asked to take a stranger's post at a model's word.