Facebook is rolling out Your algorithm, an AI-powered tool for describing desired topics in ordinary language. In its October 13 announcement, Meta says users can add and manage interests for Feed and Reels through the three-dot menu. It also says feedback after tapping Interested or Not interested may include follow-up questions, and that recommendations can respond within the same session.
Meta says large language models interpret the interests, context and intent in free-form text, combining it with existing behavioral signals. That is a company description of the system, not an independent measurement of the feature’s effectiveness. TechCrunch’s report places the Facebook launch alongside earlier Your Algorithm features on Instagram and a related Dear Algo feature on Threads.
Expressing a preference is different from defining a rule
The useful distinction is between telling a recommender what you prefer and setting a guaranteed filter. Meta’s announcement describes a way to influence ranking. It does not provide a numerical weight for a typed preference, promise that a disliked topic will disappear entirely, or explain how a request interacts with every other signal. Those missing details limit what can be concluded from the launch.
A request for fewer sports clips, for example, leaves several possible interpretations: fewer clips overall, fewer clips about a particular sport, or fewer clips from one account. A language interface gives a person more room to explain the intended preference than a binary response does. Whether the system interprets that explanation as intended is a separate question from whether it accepts the text.
That makes the wording of feedback part of the interaction. A specific request about a topic is easier to compare with the resulting recommendations than a request such as make my feed better. The latter could mean fewer posts, different subjects, different sources or a different mix of formats. This is a suggested way to evaluate the feature, not a claim that Meta requires any particular phrasing.
Look at the response, not just the input box
For someone who receives the feature, a small trial can focus on one preference at a time. Note the topic being requested, which surface is being used and what appears afterward. Feed and Reels should not automatically be treated as the same observation: a visible change in one is evidence about that interaction, not proof of an identical change everywhere on the platform.
If an unwanted recommendation still appears, that observation would not by itself show that the feature is broken. A ranking influence and an absolute exclusion have different success conditions. Conversely, seeing several relevant posts after sending feedback would not establish that the prompt alone caused the change. A short viewing session cannot isolate all the factors that may have shaped the recommendations.
A useful record would therefore describe what happened rather than invent a performance score. For example: the requested subject, whether follow-up feedback was offered, and which subsequent posts appeared related to the request. These are proposed observation criteria. We have not run such a trial on a Facebook account, measured a before-and-after result or verified Meta’s same-session responsiveness claim.
What the announcement leaves open
The retrieved materials do not give a complete region-and-language rollout schedule or a detailed account of how long a typed preference remains influential. They also do not establish a satisfaction improvement or a reduction in unwanted content at population scale. Access on one account would not prove that every account has received the tool; missing access would need to be distinguished from a failed interaction after access is granted.
For a product team evaluating a similar interface, the announcement raises a concrete design question: how should a product explain the strength of a preference? A label that promises influence sets a different expectation from a label that promises exclusion. Clear feedback about what was accepted could help a user understand the interaction, but the supplied evidence does not establish which confirmation behavior Facebook provides in every case.
For creators, the evidence is narrower still. The existence of typed preferences does not demonstrate a particular effect on reach, revenue or audience distribution. Claims about those outcomes would require their own observations. The launch should be understood first as a new user input for recommendations, rather than as a proven change in the economics of publishing on Facebook.
The immediate change is a more explicit way to communicate preferences to a ranking system. Meta’s description explains what the input is meant to do. The practical value for an individual depends on the recommendations that follow and on whether the person understands the limits of that control. A conversational interface can make feedback more expressive while still leaving the underlying ranking decisions partly unexplained.
If Your algorithm is available on your account, try one specific topic preference and inspect the resulting Feed or Reels without treating it as a guaranteed filter.
