Facebook has spent two decades inferring what people want to see from what they click, watch and linger on. This week it began asking them, in writing.
Meta said it is rolling out a set of AI-powered tools that let users describe their interests in their own words and adjust what Facebook shows in the main feed and in Reels. A feature called Your algorithm can be reached from the three-dot menu on any post, and it opens a space where a user can add and manage topics directly. Ranking systems adjust as the description changes, the company said.
The feedback path runs through the same menu. Tapping Interested or Not interested can now trigger follow-up questions asking for more detail about what a user liked or disliked. The answers are free text, and Meta said recommendations update in real time, adapting within a single session to reflect what the user says matters at that moment.
The technical change underneath is what makes the feature possible. Meta’s ranking systems have long leaned on engagement signals, including likes, comments and watch time. Adding large language models lets the company read the intent inside a sentence rather than treat feedback as a category. In its announcement, Meta described the result as an algorithm you can actually talk to.
The scale of the input is the point. Meta’s ranking systems already weigh engagement, post-level feedback, and what the company describes as thousands of other signals. A written instruction is one more, and its influence depends on how much the system weighs a sentence against the record of what a person actually does.
For users, the offer is straightforward: less guessing about why a feed looks the way it does, and a way to change it without waiting for a pattern of behavior to accumulate. The three-dot menu already carried options to hide a post or snooze a person. The new tools replace a set of blunt controls with something closer to a conversation.
This is not the first version of the idea. Instagram added a comparable feature to its main feed earlier this year, and Threads introduced its own under the name Dear Algo. Facebook, which reaches a far larger and more varied audience, is the harder test, partly because its feed mixes friends, groups, pages and advertising in a way the other apps do not.
Regulation sits behind the timing. European rules require the largest platforms to offer users a version of their feed that is not based on profiling, and they oblige the companies to explain how ranking works. A feature that lets users state preferences in plain language addresses both obligations without dismantling the ranking system that produces engagement.
That is where the commercial tension begins. Meta sells attention, and the ranking layer is the machine that manufactures it. Inviting users to redirect the machine is a concession that could trim the time spent in the app, the metric the advertising business is built on. Meta’s answer is that relevance and revenue move together, and that users who see what they asked for are less likely to drift away.
Advertisers will be watching for second-order effects. If prompts change what appears in the organic feed, they also change the context in which paid posts are shown, and the tools give users a way to signal preferences that media buyers currently infer from panels and surveys. None of that has been priced or measured in public.
There is also the question of durable control. A prompt entered in October describes what a user wants that day. The system adapts within the session, but Meta’s announcement did not say how long a stated preference outlasts the moment it was given, how quickly an old instruction decays, or whether an explicit request can override a strong behavioral signal pointing the other way.
A stated preference is also a record. Whatever a user types becomes an input the company stores and applies, and Meta’s announcement did not describe how long those inputs are kept or whether they inform anything beyond the feed.
The company also did not say how many users would receive the tools, or how it will judge whether they work. The feature repackages a long-running argument between platforms that design feeds around inferred behavior and users who want a say in what they see. This time the platform is handing over the pen and keeping the ability to edit.
For the person who types an instruction and watches the feed change, the question of measurement will not come up at first. The visible promise is that the algorithm takes direction. The hidden one is how much of the direction it keeps once the session ends, and who decides when a request and a habit disagree.
Meta has spent years arguing that its systems reflect user intent rather than shape it, an argument it makes to advertisers, regulators and users at the same time. A text box is the clearest evidence it has offered for that claim, and the easiest for outsiders to test. What the prompts reveal about how the feed actually responds will be read well beyond the people typing them.
