Facebook introduced a new "Your Algorithm" feature that lets users directly add and manage topics they want to see in their feed and Reels. Meta says recommendations respond right away after users mark topics as "interested" or "not interested." The feature previously launched on Instagram, and a similar "Dear Algo" feature exists on Threads.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Meta is expanding Instagram's "Your Algorithm" feature to Facebook and says it is using large language models more in its ranking systems, allowing recommendations to respond in real time to users' natural language expressions within the same session. But what "more use" actually means is something the report does not quantify in detail.
Meta is moving "Your Algorithm" from Instagram to Facebook. Facebook users can now directly add and manage the topics they want to see in their feed and in Facebook Reels. This is not an isolated launch: the feature previously went live on Instagram earlier this year, and Meta also rolled out a similar "Dear Algo" feature on Threads. Three platforms, three names, all pointing to the same product logic—putting part of the control over the recommendation algorithm into users' hands.
The entry-point design continues this "low-barrier" approach. Facebook users can access "Your Algorithm" through the three-dot menu, or select "Interested" or "Not interested" on an individual post; these two buttons may also appear below posts. In other words, users can manage topics in a centralized settings interface, or give feedback on the fly when they come across specific content.
Meta's description of this feature centers on natural language. Meta says users can tell the system in their own words what they want to see more or less of, and that recommendations update in real time; after making an update using the new feature, recommendations will "respond right away." Meta also says it "may ask follow-up questions" to learn more about what users like or dislike.
These statements all come from Meta itself and are source_claim rather than confirmed fact. The report provides no measured data beyond interface screenshots, and does not specify the actual latency scale of "real time." Based on the report, Meta did not explain the specific scope or proportion of large language model use in its ranking systems, so "more use" lacks quantifiable detail.
Meta says it is using large language models more in its ranking systems, and says large language models can incorporate the interests, context, and intent behind natural language, thereby "unlocking an algorithm you can actually have a conversation with." Meta also says recommendations will adapt within the same session to reflect what users care about at that moment.
This is the most technically significant and also the least detailed part of the entire report. "More use" is a relative statement—no baseline, no proportion, no specific modules. Whether large language models are used to understand user input, generate candidate content, or participate in final ranking, the report does not say. Therefore, for the phrase "conversational algorithm," readers can currently only treat it as a statement of Meta's product vision, not a verifiable technical description.
The new feature is part of what Meta calls a set of "AI-powered" tools for customizing what Facebook users see. This positioning places "Your Algorithm" within a larger product narrative: not a single feature, but a collection of tools that let users intervene in their feed.
But "AI-powered" is itself a broad label. The report does not list the specific tools in this set, nor does it explain how the tools work together. What can be confirmed is the existence of the feature and its entry points; what cannot be confirmed is the extent to which this set of tools changes the actual ranking results of the Facebook feed.
Looking at confirmed facts and source claims separately, the boundary is fairly clear. What is confirmed: Facebook launched "Your Algorithm," allowing users to directly add and manage the topics they want to see in their feed and Reels; the feature previously launched on Instagram; Meta also rolled out "Dear Algo" on Threads; users can access it through the three-dot menu, or select "Interested" or "Not interested" on an individual post; these two buttons may also appear below posts.
What belongs to Meta's claims: it may ask follow-up questions; users can express preferences in their own words and recommendations update in real time; after making an update, recommendations "respond right away"; it is using large language models more in its ranking systems; large language models can incorporate the interests, context, and intent behind natural language; recommendations adapt within the same session. None of these have been independently verified.
The report only relays Meta's claims and provides no independent third-party verification of the "real-time update" or "same-session adaptation" effects. This means that, regarding how this feature actually performs, there is currently no cross-checkable evidence in publicly available information.
All information in this article comes from a single report by The Verge, which itself primarily relays Meta's official statements. Confirmed facts are limited to the feature's existence, launch platforms, and entry methods; content involving effects, real-time performance, and the extent of large language model use are all unilateral claims by Meta, and the report provides no third-party verification or quantitative data. Readers should treat phrases such as "conversational algorithm" and "same-session adaptation" as product promises rather than verified conclusions.
Meta's expansion of "Your Algorithm" to Facebook marks the start of mainstream social platforms experimenting with natural language interaction to let users directly intervene in recommendation results. But "more use of large language models" lacks quantifiable detail, and the performance claims are backed only by Meta unilaterally. Whether this feature can truly make the algorithm "understand plain speech" remains a product promise awaiting verification, not a proven capability.
Primary report
Primary source