Summary: Meta has started rolling out “Your Algorithm” on Facebook, letting users describe in natural language what they want to see in the feed and Reels. Per TechOrange’s coverage of the announcement, the LLM-backed ranking system responds in real time, adjusting recommendations within the same session. For product builders, it’s a rare example of conversational controls replacing opaque ranking signals in a mainstream consumer surface.
What the feature actually does
Users can add and manage topics they want to see in their feed and Reels. Meta also updated the three-dot menu on each post: from there you can open “Your Algorithm,” or use “Interested” and “Not interested” buttons — which may also appear directly under posts. When you tap one, the platform may ask why, and you can explain in your own words.
The interesting technical shift is underneath. Facebook’s ranking has always inferred interests from thousands of signals — likes, comments, watch behavior. Meta says it’s now using large language models more broadly in that ranking system, describing it as an algorithm you can genuinely talk to: state what you want more or less of, and results respond immediately, within the same session.
The rollout is gradual, and so is the disclosure
Instagram got “Your Algorithm” first, in Reels in December 2025, with feed and Explore added in June 2026. Threads shipped a similar feature called “Dear Algo” in February 2026 across the US, UK, Australia, and New Zealand, with adjustments lasting 72 hours. Facebook’s version is rolling out now, but as PPC Land pointed out, Meta hasn’t said when it will be fully available, which countries get it first, how user input weighs against behavioral signals, whether added topics persist or decay, or whether inputs feed ads, storage, or model training.
What it means for reach and for product design
For page owners, these stated preferences are signals they can’t see — the effect only surfaces as changes in reach. Since feed and Reels carry most of Meta’s ad impressions, organic ranking shifts could ripple into ad delivery too.
The design pattern is the notable part for builders: instead of forcing users through settings toggles, the product accepts an intent statement and maps it onto ranking. It’s the same instinct behind exposing model or agent behavior through language rather than configuration — something I touched on in the earlier piece on querying agent traces in plain English, where natural language becomes the interface to a system’s internals.
The open questions are the ones worth tracking: how long a stated preference lasts, and whether your words shape the ads you see. TechCrunch’s read is that Facebook is moving away from the traditional recommendation model toward something closer to a streaming service’s taste-driven customization — but with far less visibility into the tradeoffs than a streaming queue gives you.
A practical next step if you run pages on Facebook: watch reach deltas on unaffected content over the next few months. If your distribution moves without a visible cause, stated user preferences may now be part of the ranking mix.