Meta rolls out AI-driven controls that let Facebook users steer Feed and Reels in plain language

Meta is introducing a set of AI-powered tools on Facebook designed to put users in control of what they see in Feed and Reels. In a newsroom post on October 13, the company said people will be able to shape recommendations in “your own words,” and see the results respond right away rather than after repeated browsing or engagement.

The update centers on new ways to customize recommendations based on natural language. Meta said its ranking systems already infer what someone is interested in from engagement indicates, including posts people like, comment on, and watch. With the new tools, the company says large language models (LLMs) can use the meaning, context, and intent behind a user’s text to adjust what rises in Feed and Reels, adapting during the same session.

Meta also described a feature called “Your algorithm,” which gives users a dedicated place to add and manage topics they want to see. The company said people can find it through the three-dot menu on the Facebook interface, positioning the controls as something accessible from individual posts rather than buried in settings.

For feedback, Meta is expanding how users can react to content they do or do not want. The company said it is adding more ways to provide feedback directly from posts in Feed and Reels, using “Interested” and “Not interested” actions. After clicking, Meta said it may prompt follow-up questions to clarify what the user liked or disliked, including through freeform answers in the user’s own words.

Meta framed the change as a conversational approach to ranking. Instead of treating recommendations purely as an automatic output of past behavior, the company said LLMs allow the system to interpret written preferences and connect them to the underlying ranking logic that decides which items appear.

Under the hood, Meta said it is evolving how ranking works by combining freeform text with everything the company already understands about what a person enjoys. The company’s stated goal is for recommendations to be “more relevant, more nuanced, and more you,” while continuously improving the system’s responsiveness over time.

Meta also said it updated the three-dot menu on every post to make access to the new customization tools easier, suggesting the company expects users to encounter the controls in the normal flow of using Facebook rather than as a separate feature hunt.

The company’s broader bet is that better recommendation relevance depends not only on how people interact with content but also on how clearly they can express their intent. LLMs can interpret natural language statements such as “more of this topic” or “less of that kind of content,” and Meta said the system can incorporate that intent into ranking decisions quickly.

Even with Meta’s description, key details remain unclear from the announcement alone. The company did not specify which regions or account types will receive the feature first, how long the “same session” response typically takes in practice, or whether the tools affect both Facebook Feed and Reels equally. It also did not outline what happens when users provide conflicting preferences or how the system prioritizes explicit “your words” inputs versus engagement history.

Still, the direction is clear: Meta is moving toward more interactive recommendation control, turning feedback and preferences into a more direct input channel for its ranking systems. For users and advertisers watching platform behavior, the next question will be how quickly these tools roll out and whether they measurably change the composition of content people see over time based on explicit, language-based steering rather than clicks and watches alone.