Meta wants you to talk to the Facebook algorithm. On October 13, 2026, Meta announced that Facebook is getting "Your Algorithm," a control that lets you add and manage the topics you want more or less of in your feed and in Reels, according to The Verge's report on the launch. Meta also says it is leaning on large language models in its ranking systems so those requests land quickly.
It is a modest-looking feature with a bigger claim behind it: that ranking can move from hidden signals toward something closer to instructions. This guide covers what Meta has said, how it fits the Instagram and Threads versions, what to be skeptical about, and how to use the controls today.
TL;DR: Your Algorithm on Facebook
| Question | Answer |
|---|---|
| What is it? | A topic-level control for your Facebook feed and Reels |
| How do you use it? | Open Your Algorithm, or choose interested or not interested on a post |
| Natural language? | Yes: Meta says you can say in your own words what you want more or less of |
| How fast? | Meta says recommendations respond right away, within the same session |
| Where else? | Instagram (earlier in 2026) and a similar Dear Algo on Threads |
| Does it use LLMs? | Meta says it is using LLMs more in ranking; the share is not disclosed |
| Rollout details | Countries and timing were not spelled out in the report we read |
What Meta says it launched
The Verge quotes Meta describing the feature as part of "AI-powered" tools to customize what people see. Three mechanics are described. First, you can directly add and manage topics for the feed and Reels. Second, you can hit interested or not interested from the three dots menu on a post, and those buttons may also appear beneath posts. Third, Meta "may ask follow up questions for more detail on what you liked or disliked."
The company framed the technical change this way: "LLMs can incorporate the interests, context, and intent behind natural language, unlocking an algorithm you can actually talk to." That sentence is the interesting one. A classic recommender learns from clicks, watch time and shares. A language model can read a sentence like "less politics, more woodworking and local hiking" and map it onto content it has never seen you touch.
A river of content cards with one card selected, illustrating how a Facebook feed ranks posts and how Your Algorithm steers it
Where this came from: Instagram and Threads
Facebook is the third surface. Meta tested Your Algorithm on Instagram Reels in late 2025 and later extended it to English-speaking users globally, as Social Media Today reported. The Instagram version lets you add or remove topics by hand. Meta's own transparency page on user controls lists Your Algorithm and Dear Algo under "Customize your feed", separate from the older interested and not interested controls.
On Threads, Meta has a similar feature called Dear Algo, which The Verge describes only as "similar". The through-line is that Meta is testing manual algorithm inputs on all three apps at once and now carrying lessons to Facebook, where the audience is older and the feed is more mixed between friends, groups, pages and recommendations.
Why LLMs in ranking is the real story
Meta's executives have been cautious in public about this shift. Coverage of CFO Susan Li's March 4, 2026 remarks at a Morgan Stanley conference reported her saying LLMs were not yet a large part of core ranking, with compute capacity a constraint and Threads among the places where they were used in limited ways. Meta's engineering blog has described LLM-scale models for ads in its Adaptive Ranking Model post, but that is advertising, not the organic feed.
So the new Facebook statement is notable mostly as a direction of travel. What we can say from sources: Meta is publicly claiming more LLM use in feed ranking. What we cannot say: how many ranking decisions involve an LLM, which model, or whether the language model sits in the main ranker or in a lighter layer that interprets your preferences and feeds them to a conventional system. The second design is far cheaper, and it would match the "respond right away" claim.
Why this matters beyond Facebook
Three reasons readers who build or study AI should care.
Preference as text. Natural-language steering of a recommender is a product pattern other platforms will copy. If you build recommendation features, the lesson is that users prefer a sentence over a settings page, and that the hard part is turning that sentence into a stable ranking signal.
Evaluation gets harder. When a user says "less drama," what counts as success? You need tests that check whether the feed actually changed, not just whether a model parsed the sentence. Our look at production agent trace scoring in LangSmith covers a similar problem for agents.
AI slop pressure. Feed controls arrive while platforms face complaints about synthetic and low-quality posts. Topic muting does not detect AI-generated media. Tools like Google's SynthID verification site address provenance, which is a different layer.
What people are likely to ask next
Will creators see a difference? Probably, though Meta has not said so. If viewers can mute whole topics, pages that post across many themes may lose reach among people who muted one of them, while tightly focused pages could gain from people who add their topic. Creators should watch their audience retention by topic rather than reach alone.
Is this a response to complaints about the feed? Meta has not tied the launch to any complaint. Giving users a say is also a cheaper answer than rewriting the ranker.
Can I opt out of the AI parts? The report does not describe an opt-out. You can simply ignore the feature, and the ordinary feed keeps working with your usual signals.
Does it work for friends and groups content? The announcement talks about the feed and Reels. Whether topic controls affect posts from friends and groups, as opposed to recommended content, was not clarified in the coverage we read.
How to use the controls today
- Open Facebook and look for Your Algorithm in the feed or Reels settings. The Verge says it is also reachable from the three dots menu on a post.
- Add topics you want more of and mute topics you want less of. Be specific: "sourdough baking" will steer better than "food".
- Use interested or not interested on individual posts for fast signals, and answer follow-up questions if Meta asks for detail.
- Check back after a day. The claim is that changes apply in the same session, but real feeds are noisy and one session proves little.
- Review your ad preferences and activity settings if you are uneasy about preference data being used for targeting.
What to be skeptical about
- No independent test yet. The "respond right away" line is Meta's. We have not seen measurements of how quickly or how well topic changes shift a feed.
- Engagement incentives. A platform that earns from attention is also the party deciding how strongly to honor your request. Control and optimization can pull in different directions.
- Scope. The report does not list countries, languages or account types, so assume a staged rollout.
- Data. Telling an algorithm what you like is an extra signal for Meta. The coverage does not say whether it feeds advertising.
- Not a safety tool. Muting a topic does not remove scams, deepfakes or harassment. Use the report and block tools for those.
How this fits Meta's wider AI push
Meta is using AI across consumer surfaces at once: the Muse assistant family, the dossier and privacy debates around it, and now the feed. For context, see our coverage of what Meta got right with Muse, the Muse daily user plateau, the Muse dossier investigation and the Verge analysis of AI agent privacy promises. Together they show a company trying to make AI feel like a personal layer over its apps, while questions about data use follow each launch.
Concentric rings reflecting the same dot, a picture of the filter bubble risk when a Facebook algorithm is tuned to your stated preferences
One more design risk deserves a mention: steering a feed toward only what you say you like can narrow it. People often claim to want less of things they still engage with, and a model that obeys literally may produce a quieter, more insulated feed. That is a user choice, but it is worth making deliberately.
Bottom line
Your Algorithm on Facebook is a sensible, low-risk feature that gives users a clearer handle on their feed, and it is the first time Meta has tied that handle publicly to LLM-based ranking on Facebook. The confirmed parts are the controls and Meta's own description. The unproven parts are the speed, the depth of LLM involvement and the effect on slop, all of which need outside testing.
Details reflect The Verge's report and Meta's public pages as of October 13, 2026 and may change as the rollout expands.