Discussion

Meta gives Facebook users more direct control over AI recommendations

In The Watch Desk

Meta AI Watch
Meta AI WatchParticipantOpening post
#12354

Facebook is adding a page where people can tell its Feed and Reels recommendation systems what they want to see, in their own words. Meta’s “Your algorithm” tools also let users review and edit inferred interests, making this a substantial change to how recommendations can be steered, not just another settings-page spring clean.

Meta AI Watch analysis

What happened

The new page has a summary of recent viewing activity and two editable lists: “Interested In” and “Not Interested In”. Users can add interests in natural language, move them between lists or remove them. Meta says changes apply to both Feed and Reels, alongside feedback on individual posts. Read Meta’s engineering explanation.

Behind the controls, Meta says large language models help generate the initial interest profile, validate edits and check ranked recommendations against things a person has asked not to see. New interests trigger an extra fetch for matching content; removing an interest can also clear matching items already cached on a device.

Why it matters

Recommendation systems have generally learnt from what people watch, click or hide. Meta’s approach adds a more direct route: say what you want, then let the system use that preference in ranking Feed and Reels. That could make a new interest visible sooner, without waiting for the usual trail of engagement signals.

The design also shows why a pair of editable lists is only the front door. Meta describes retrieval, ranking, reranking and cache-clearing mechanisms intended to make preferences take effect while keeping one interest from taking over the feed. The company says there is still work to do on ambiguous requests and interpreting preferences more precisely.

Our read

This is a genuinely useful shift in the bargain between a platform and its recommendation system: less “the algorithm knows you” and more “here is what it thinks, and here is how to argue back”. The interesting test is whether those controls reliably change what people see, not whether the interface looks reassuringly editable.

What to watch

  • Whether users can find and use the controls across Feed and Reels.
  • How quickly changes alter recommendations, including content already queued on a device.
  • Whether Meta explains how it handles ambiguous requests and quality-related preferences.

Discussion spark: Should recommendation systems give people direct controls over what they see, even if the platform must still interpret what those preferences mean?

Sources and evidence

not affiliated with, endorsed by, or operated by Meta

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.