Meta launches Your Algorithm on Facebook with natural-language feed control

Meta announced yesterday that it is rolling out "Your Algorithm" on Facebook, a feature that lets users steer recommendations by writing what they want in plain language instead of relying solely on passive signals. The tool first appeared on Instagram Reels in a limited pilot last December, expanded to the full Instagram experience in June 2026, and is now arriving on the Facebook feed and Reels with a promise that ranking updates take effect in real time after every change.

In practice, a user types a free-form description — "less political content, more cooking tutorials" — and the system translates that into ranking weights. According to Meta, the new weights are reflected in recommendations immediately, without waiting for a model retraining or a daily batch update. The three-dot menu on individual posts has also been updated so users can mark "Interested" or "Not interested" directly, and the platform may ask follow-up questions to clarify intent.

Classic recommendation systems learn from clicks, watch time and other indirect interactions, creating feedback loops that tend to amplify extreme or addictive content. Meta is attempting to mimic the streaming-service model: the user declares a preference and the system implements it. The open question is how much weight the explicit declaration receives relative to existing behavioral signals, and whether the mechanism can handle contradictions between what people say they want and what they actually consume.

Across Meta's ecosystem, Instagram received the full version in June 2026 after the December 2025 Reels pilot. Threads operates a parallel feature called "Dear Algo" that works on the same free-text principle. The three platforms share parts of a common ranking infrastructure, so lessons from one are meant to transfer to the others, but Meta has not released comparative performance metrics — click-through rate, dwell time, retention — that would show whether the new approach genuinely improves relevance or merely creates a feeling of control.

The announcement leaves several gaps. It does not explain how the system handles contradictory requests, whether there is a length limit on the free-text field, or how abuse is prevented — for example, users stuffing keywords to promote specific content. It is also unclear whether the weights a user sets are visible to them or remain a black box. Without independent data or researcher access, "Your Algorithm" remains a product promise rather than a proven result.