Facebook now has its own AI-powered search mode, and it works very differently from Google. Ask it where to find the best hiking trail near you, which local restaurant people actually recommend, or what parents are saying about a school in your area, and instead of a list of blue links, you get a synthesized answer drawn from what real Facebook users have been posting, debating, and recommending in Groups and Reels. For anyone who has spent the last decade typing questions into Google only to land on an SEO-stuffed article from a site they’ve never heard of, the appeal of that is obvious. Whether it delivers is a different question.
On June 15, 2026, Meta Platforms announced the launch of a new feature called AI Mode on Facebook. The rollout is part of a broader push to make the platform feel less like a place you scroll out of habit and more like somewhere you actually go to get things done. The update arrives at a moment when Meta is in an unusually aggressive phase, spinning up new products, building proprietary AI from the ground up, and watching its stock sit above $700 a share.
AI Mode changes how Facebook handles search. Instead of returning a conventional set of links or posts, it produces a synthesized answer built from what people are saying publicly across Meta’s apps. Your question goes in. A synthesized response comes out. No scrolling, no picking between results. Just an answer, constructed from the collective public output of Facebook’s enormous user base. The description sounds simple. The implications are not.
What AI Mode Actually Does
Meta has officially introduced Facebook AI Mode for US users, transforming the standard search bar into a conversational tool that answers questions by mining public Group discussions, Reels, and Marketplace data. The feature appears as a toggle inside the existing Facebook search interface, so there’s no new app to download, no separate surface to find. It’s built into the search bar most people have been ignoring for years.
As Meta explains it: “AI Mode uses Meta AI to give you answers grounded in what people are saying publicly across our apps like in Groups and Reels, so you get real perspectives and experiences rather than a generic list of search results.” That framing is deliberate. The pitch is authenticity: lived experience from real people in real communities, not whatever managed to rank highest on a search engine.
Meta’s own product page states that answers and recommendations can “cite public posts from Instagram, Facebook, and Threads” to provide “richer responses” rooted in what people are talking about now. That cross-platform pull is significant. It means a question asked on Facebook can draw on conversations happening on Instagram or Threads, creating a synthesized answer that aggregates public opinion from across Meta’s entire ecosystem.
AI Mode is live on the Facebook mobile app for US users, with no confirmed timeline for a desktop version or international rollout. A UK launch timeline and any web or desktop version have not been confirmed by Meta.
The Model Behind It: Muse Spark
This feature is powered by Muse Spark, designed to provide users with quick answers to their queries directly on the platform. That name is worth understanding, because Muse Spark is not a minor upgrade to a previous model. On April 8, 2026, Meta announced Muse Spark, the first in a new series of large language models built by Meta Superintelligence Labs, whose team spent the preceding nine months rebuilding Meta’s AI stack from the ground up.
In the past, Meta was caught manipulating the published benchmark results of an AI model to make it appear more capable than the version available to most users actually was. Developer trust collapsed. Mark Zuckerberg changed strategy entirely. In June 2025, Meta spent $14.3 billion to acquire a 49% nonvoting stake in Scale AI and brought in its cofounder and CEO, Alexandr Wang, as Meta’s first-ever chief AI officer, tasking him with leading the newly created Meta Superintelligence Labs.
Muse Spark is the result. According to the Artificial Analysis Intelligence Index, Muse Spark achieved a score of 52, placing it fourth overall behind Gemini 3.1 Pro Preview, GPT-5.4, and Claude Opus 4.6. Meta’s previous flagship, Llama 4 Maverick, debuted in 2025 with an Index score of just 18. For a company that spent much of 2025 being mocked for benchmark manipulation, that’s a significant turnaround. And it’s the engine now powering the AI answers that hundreds of millions of Facebook users are about to start receiving.
Forum: The App That Arrived First
AI Mode didn’t emerge from nowhere. Meta quietly dropped a new app on May 22, 2026, that looks a lot like Reddit, with no announcement, no press event, just a listing in the Apple App Store. Forum is a Reddit-style app that includes its own AI “Ask” tab, letting users pose questions and get answers pulled from discussions happening across Facebook Groups.
Forum and Facebook are still linked, meaning you can enter Forum with your Facebook login, and whatever you post there will be visible in your groups on the Facebook app as well. There is also an additional AI assistant for group admins, which assists them with tasks like content moderation.
Reddit shares fell almost 6% on Friday on concern that Forum could create an alternative avenue for internet users to congregate and create discussion groups. Truist analysts wrote in a note that “the risk from this move, if successful, is a gradual erosion of Reddit’s utility for casual users who have less community loyalty to Reddit and simply want answers.” This is Meta’s second attempt at a standalone Groups app: the first launched in 2014 and was killed in 2017. The difference this time is the AI layer, and the company’s vastly larger infrastructure for building products at speed.
Seen together, Forum and AI Mode form a coherent strategy. Forum is a standalone product for people who want a dedicated community discussion experience. AI Mode takes the same underlying logic – synthesize answers from real human conversations in Groups – and bakes it directly into Facebook’s existing search. Same idea, two surfaces, covering different user behaviors.
The Reliability Problem
The obvious question, and it’s a fair one, is whether any of this can actually be trusted. Because the AI is summarizing content from everyday users rather than vetted sources, there’s a real risk of outdated or misleading information slipping through, a concern that’s already been raised about Google’s own AI Mode on Reddit.
Facebook Groups contain multitudes. There are highly informed communities where specialist knowledge is shared freely and carefully, and there are communities where misinformation travels fast and corrections travel slow. The new chatbot feature synthesizes responses from Facebook Groups and Reels, but critics warn that user-generated content isn’t a reliable source, and Meta has not publicly responded to these accuracy concerns.
When you type a question into Forum’s Ask tab, the AI pulls together relevant answers from existing conversations and posts across Facebook Groups. Users also get links to those threads to dig deeper or double-check the information. That link-back feature matters. It gives users a route to the original source and, in theory, a way to assess whether the synthesized answer is actually representative of the discussion underneath it. Most people will likely read the summary and act on it; whether they click through to verify is another matter.
The Privacy Dimension
There’s a data layer to all of this that’s worth being clear about. Meta began training its AI systems on EU and EEA user data from May 27, 2025, after pausing in 2024 over GDPR consent concerns. US users fall outside those protections, with their data use for AI training governed by Meta’s standard terms.
In Europe, the picture is complicated. As of May 27, 2025, Meta started using some of the personal data of European users to train its AI systems, including tools such as Meta AI and Llama language models. Meta initially announced this initiative in 2024 but paused its implementation following discussions with the Irish Data Protection Commission, which raised concerns about the legal basis and transparency of the proposed data use.
Meta will use public posts, comments, and interactions with Meta AI across its platforms to train its generative AI systems. The company has stated that no private messages or content from users under the age of 18 will be included. Users in the EU can opt out by submitting an objection form, aligning with GDPR’s emphasis on consent and transparency. Whether an opt-out mechanism is a sufficient safeguard, or whether something closer to explicit opt-in consent is required, remains actively contested in European courts and regulatory bodies.
The Bigger Picture: A Platform Trying to Stay Relevant
Step back from the individual features and a broader pattern is visible. Meta wants Facebook’s AI tools to make the platform stickier and more useful, while also diversifying how it makes money. Alongside these feature rollouts, the company recently launched global subscription plans for Facebook, Instagram, and WhatsApp starting at $3.99 a month, with more AI-related subscription tiers reportedly on the way.
Meta’s broader app strategy signals that CEO Mark Zuckerberg told employees that with AI-driven efficiencies allowing the company to build more apps, the social media giant now aims to roll out many more apps than it has historically.
AI Mode fits Meta’s broader product strategy: keep users inside Facebook longer by making search, creation, and editing happen in one place. For Meta, the immediate benefit is less about a new hardware line or a standalone subscription and more about turning an old app into a more active AI surface.
AI Mode, in that context, is not just a search feature. It’s an argument that Facebook is still worth opening. That the 20-year-old platform where your relatives argue about politics and share photos from weddings you weren’t invited to is also, somehow, the right place to get a reliable recommendation for a plumber or a pediatrician or a hiking trail. Whether people believe that depends on whether the answers AI Mode gives them turn out to be good ones.
Read More: Why Mark Zuckerberg Warns Against Screenshotting Facebook Messenger Chats
What to Make of All This
The most honest way to read this launch is as a bet. Meta is betting that the sheer volume of real human conversation it holds, the kind you can’t find on any other platform at that scale, is more valuable as an AI training resource than as a social feed. The Groups where parents swap advice about schools, where locals debate restaurants, where hobbyists share niche knowledge: all of it becomes source material. All of it gets synthesized into an answer.
That separates AI Mode from ChatGPT or Google Search in a meaningful way. Those tools pull from the indexed web. Meta’s AI Mode pulls from conversations that were never meant to be search results. Raw, unmoderated authenticity could be a feature. It could also be a liability, and there’s no way to know which until enough people have used it long enough to find out.
Some of these concerns about accuracy and data use are not going to resolve quickly. The GDPR case in Europe is still working through the courts. Misinformation has been a persistent problem in Facebook Groups for years, and powering an AI answer engine from the same source material doesn’t make those problems smaller. A bad Group post was once only seen by the people who follow that group. Now it might become part of a synthesized answer surfaced to anyone who asks the right question. That’s not a hypothetical risk. It’s a design consequence worth sitting with.
AI Disclaimer: This article was created with the assistance of AI tools and reviewed by a human editor.