← All writing
16 Nov 2026 · Meta

Meta's Your algorithm lets you steer your Facebook feed

Most of us have scrolled a social media feed and wondered why it keeps showing us the same sort of thing. Meta has described a new set of AI-powered tools on Facebook, called Your algorithm. It lets people say, in their own words, what they want to see more or less of in Feed and Reels.

The details come from a post on Meta's engineering blog, which explains what the tools do and how they work. Some things it leaves unclear.

What problem is it trying to solve?

Until now, the systems that choose what you see on Facebook have mostly learned by watching what you do. If you pause on a video or like a post, that counts as a hint. Meta calls these hints 'implicit engagement signals'.

Meta says this approach works well for huge numbers of people, but it adapts slowly and misses nuance. It is a bit like a waiter who only learns your tastes by watching which plates come back empty, when he could just ask you what you fancy.

Your algorithm is Meta's attempt to let people do the telling. In Meta's words, the aim is to make recommendations more transparent and easier to steer.

What does the page look like?

The Your algorithm page has three parts:

  • A one-sentence 'Lately Summary' at the top, describing what you have been watching recently.
  • An 'Interested In' list, already filled in with interests Facebook has guessed from your past activity.
  • A 'Not Interested In' list, holding topics you have marked as not interesting, or posts you have crossed out or hidden.

You can add new interests by typing them in everyday language. You can also move them between the two lists, or remove them. Meta says every change applies to both Feed and Reels, and keeps updating as you react to individual posts.

How do your words change what you see?

When you add something to 'Interested In', Facebook goes looking for it on purpose. Each interest gets four dedicated ways of finding matching posts, including searching and looking at closely related topics. Without this, Meta says, posts about a brand-new interest might never be considered at all.

Meta wants a new interest to appear in your feed without crowding out everything else. The system spaces out posts on the same interest and limits how much each interest can add each day. New interests get an extra boost for a limited time. After that, Facebook learns from how you actually respond to them.

When you add something to 'Not Interested In', the system tries to block it at several stages, because topics are fuzzy and any single filter will miss some matches. First, it compares the meaning of each post with your unwanted topics and removes close matches. Then it pushes near matches further down. Finally, an AI gives the list a last check.

Some requests are not really topics. Asking for less AI-generated or sexually suggestive content is handled separately, as a matter of quality rather than subject. If you name a particular creator, their content is filtered out directly.

Changes also have to take effect quickly. Your phone often holds a queue of posts that were chosen before you made a change, so when you add an interest, Facebook fetches fresh recommendations straight away. When you block one, your phone throws out any queued posts that match it before they appear.

Where does the AI come in?

Meta uses large language models, or LLMs, at several points. An LLM is a type of AI that reads and writes text, the same kind of technology behind chatbots.

First, an LLM builds your starting profile. It turns clues from your viewing into readable topic names for the 'Interested In' list, and it writes the Lately Summary.

Second, every time you edit the page, a smaller model that is cheaper to run checks your change. It makes sure what you typed is a sensible topic, spots duplicates and tidies up the wording. Meta says this is only one layer of checking; a separate part of the system handles integrity and safety checks.

Third, a specially trained LLM reviews the final list of posts and removes anything related to topics you recently added to 'Not Interested In'. Meta found that a small model trained to name only the posts to remove was faster and more reliable than a general-purpose model judging each post one by one.

To test these AI steps before launch, Meta used another LLM as a kind of examiner, grading each answer against the behaviour Meta expected.

What is still unknown?

The engineering post focuses on how the system works. The version available to us is cut short, so it does not include detailed test results. It also does not say who can use Your algorithm, where, or when.

Meta admits there is more to do. It wants to get better at understanding vague requests and to make recommendations respond more precisely to each one. It is also exploring more places in the app where people could shape their preferences.

For small businesses, the obvious question is whether this changes who sees your posts, especially now that people can block a topic or a creator by name. Meta's post does not discuss businesses or advertising, so for now that question has no answer.


Source: Your algorithm, Meta (Engineering at Meta). Listed in AI News Week 47, where we track every AI release each week.