Facebook’s Your algorithm lets users describe topics they want more or less of, with Meta detailing today how those requests change recommendations across Feed and Reels. The system uses large language models to interpret written preferences and combines those requests with activity signals when choosing content, according to Meta. Users can edit their interests in one place, with both recommendation systems drawing from the same stored preferences.
The controls include an Interested In list populated from previous activity, plus a Not Interested In list reflecting explicit feedback and content people have hidden. Users can type additional interests, delete entries or shift topics between the lists. A short Lately Summary describes recent viewing, giving users another way to review how Facebook interprets their activity.
Read more: Meta Ordered to Pay $25 Billion Over Facebook Privacy Claims
Meta began rolling out Your algorithm in October, with access through posts’ three-dot menus and a promise that recommendations could adjust within the same session. That rollout also included follow-up questions after users selected Interested or Not interested on a post, allowing them to explain their reactions in text.
Adding an interest does not give it unlimited space: Meta says the system separates recommendations about the same interest and restricts its daily contribution. Newly entered interests temporarily receive greater weight, after which engagement with that content helps guide recommendations. Each request for recommendations retrieves candidates for one interest at a time so different interests get a turn.
For unwanted topics, Facebook combines several checks, including comparisons of topic meaning, ranking penalties for close matches and a final review by a language model. Requests to reduce AI-generated material or sexually suggestive content use dedicated signals that lower its ranking, while requests naming a creator trigger direct filtering of that creator’s content. Meta says the final model review removes items it considers related to a recently rejected interest.
The delivery system also addresses recommendations prepared before someone changes a preference. Adding a wanted topic triggers an extra request for fresh content, while adding an unwanted one causes matching material already stored on the device to be removed before display. Meta says interpreting ambiguous requests and making recommendations respond more precisely remain areas for improvement.
