Meta's new feature uses three lists to manage recommendations across Facebook Feed and Reels. The central judgment is straightforward: algorithmic recommendations are finally beginning to accept explicit user instructions rather than relying only on inferred click behavior.
What this is
Your algorithm is a recommendation-control tool Meta is rolling out to Facebook users. People can tell the system in their own words what they want to see and what they do not. The system stores those preferences in a unified profile, which both the Feed and short-video feed can use.
A summary of recently viewed content appears at the top of the page, followed by two interest lists: “Interested In” and “Not Interested In.” Users can add, remove, or reorder interests. Every adjustment changes recommendation results, while later feedback on individual pieces of content continues to update the profile.
Technically, the feature does not hand every content-selection decision to a single large model. Meta divides the work into stages: a large language model—a model capable of understanding and generating natural language—interprets the preferences users express, while other models and ranking systems validate, rank, and distribute content. This architecture is intended to reduce errors in model interpretation and ensure that stated preferences actually enter the recommendation pipeline.
Industry view
The notable shift is that recommendation platforms are turning “user control” into a product feature. For years, platforms have primarily inferred interests from indirect signals such as dwell time, likes, and clicks. Your algorithm gives users a way to correct the system explicitly, which is especially useful for people who are repeatedly served similar content or need to keep learning about new topics.
Yet the feature also exposes a practical problem: preferences users state do not necessarily mean the platform will follow them over time. Recommendation systems must also balance content supply, viewing time, advertising revenue, and community rules. The more users rely on granular controls, the more platforms must demonstrate that those settings genuinely work rather than offering only an editable interface.
Meta has also not disclosed the feature's geographic coverage, rollout scope, or the degree to which recommendations may change. For now, Your algorithm is better understood as an upgrade path for recommendation systems than evidence that Facebook has completed its personalized-recommendation overhaul. Privacy remains the critical variable. Once preferences are stored centrally, how Meta defines and uses that information will directly determine whether users trust the feature.
Impact on regular people
For enterprise IT: Businesses can apply this model of explicitly stating intent and having the system continually recalibrate to internal knowledge, customer service, and content distribution. But they first need preference records that users and employees can inspect, modify, and revoke, preventing algorithms from becoming opaque black boxes.
For individual careers: The feature suggests that future AI tools may no longer come only with default settings. Being able to explain goals in natural language and correct erroneous outputs may become a basic skill for using AI. Even so, describing requirements in plain language still requires clear boundaries and validation of results.
For consumer markets: Once users can actively adjust their interests, content brands gain audience signals that are closer to genuine intent rather than merely vague traffic. However, if platforms continue to prioritize commercial conversion, those stated preferences may still be reinterpreted in pursuit of revenue targets.