Meta Finally Explained 'Your Algorithm.' It Is a Menu, Not the Controls

Open the Reels tab on Instagram and look at the top right corner. There is a small icon that looks like two lines with hearts. Tap it and Instagram shows you a single sentence about yourself, written by a machine. In the example Instagram used at launch, as TechCrunch reported, it read: "Lately you've been into creativity, sports hype, fitness motivation, and skateboarding." Under that sentence sits a list of topics you can keep, remove, or add to by typing.

It is easy to treat that screen as a curiosity. You can even share it to your Story, a little like a year-end music recap. Fewer people stop to ask the obvious question: who wrote that list, in whose vocabulary, and what happens to the words you type into it?

On November 10, Meta answered that question in more detail than it ever has. A post on the Engineering at Meta blog, credited to eight Meta engineers, laid out how "Your algorithm" works on Facebook, where Meta announced the feature on October 13. It describes a chain of language models that write your profile, check your edits, and strip posts out of your feed after ranking. Social Media Today covered the post the next day, and much of the coverage of the feature since October has leaned on the company's own framing: people are getting a say over their feeds at last.

That framing is the thing worth questioning. Read the engineering post closely and a different picture comes out. The controls are real, but they are smaller than they look, and they are built to keep the recommendation system, not you, in charge of the vocabulary. The more interesting consequence is one almost nobody covering this has spelled out: for the people who make content on Instagram, the feature quietly changes what it takes to be found. What follows walks through the mechanism, the limits Meta itself admits, the regulatory fight this tool sits inside, and what it means for anyone trying to grow an account without gaming anything.

What happens to the list after you edit it

Start with the literal mechanism, because the engineering post is unusually specific about it. According to Meta, a large language model reads regularly refreshed signals from what you have watched and engaged with in Feed and Reels, and turns them into readable topics. Those topics prefill your "Interested In" list. The same model writes the one-line "Lately" summary at the top of the page.

When you add or move a topic, a second, smaller model steps in. Meta describes it as inexpensive to serve, and its job is narrow: check that what you typed is a coherent topic, catch duplicates, and map your phrase to a "clean topic name." Your edit is then given an embedding (a mathematical fingerprint of its meaning) plus search keywords, and saved to a single store that both Feed and Reels read from.

Each interest on your list then feeds four separate ways of finding candidate posts. Two are direct: search, and clusters of closely related topics. Two are indirect: Meta's content graph, and item-to-item similarity. Posts that pass a meaning-based relevance check against your interest get a promotion late in ranking. New interests receive extra weight for a limited period, which Meta does not define, so that you notice the change.

The "Not Interested In" list runs the same machinery in reverse, and it is stricter. Candidates whose embeddings sit too close to an entry are removed outright. Near matches are pushed down by demotion rules. Then a fine-tuned model reviews the final ranked list and deletes anything related to your recent exclusions, and the app clears matching posts already queued on your phone. If your request names a specific creator, Meta says that creator's content is simply filtered out. Requests about quality, such as asking for less AI-generated or sexually suggestive content, skip topic matching entirely and go to separate demotion signals.

Here is the image worth holding onto for the rest of this piece. Your Algorithm is a menu. Before the feature existed, the kitchen cooked whatever it guessed you would eat and never showed you what it was working from. Now it hands you a printed menu with your usual orders listed, and you can cross dishes off or write new ones in the margin. That is a genuine change. But the kitchen still printed the menu, it still decides what your handwriting means, it still sets the portion sizes, and it still watches what you actually eat.

You can edit the menu. You cannot see the kitchen.

The turn: a control that mostly answers to the system

The comparison is useful, but it undersells one thing, and the engineering post is where that thing becomes visible. Every part of the user-facing control runs through limits that Meta chose, and several of them point the same way.

The first limit is translation. You do not add a topic in your own words so much as propose one. A model decides whether your phrase counts as a topic at all, and then files it under a name from Meta's own taxonomy. If you type something oddly specific, the system decides what you meant. That is sensible engineering, since free text is messy, but it means the list on your screen is a negotiated document, and Meta holds the pen.

The second limit is the cap. Meta's own sentence for it is the most quotable line in the post: "Our approach emphasizes balance over volume: A new interest should show up, not take over." In practice, the reranking layer spaces out recommendations for the same interest and caps how much any one interest can contribute per day. Retrieval also pulls candidates for only one interest per request, so each interest takes turns. Whatever you add, the system has already decided how much of your day it is allowed to fill.

The third limit is time. The boost for a newly added interest lasts for "a limited period," after which, as Social Media Today summarized it, the topic is weighted by how you interact with it, like everything else. In other words, the tool can push the feed toward what you say you want, but your behavior gets the last word. If you add "classical guitar" and keep scrolling past it, the system will learn the scroll, not the wish.

The fourth limit is measurement, and Meta is candid about it. "Standard offline metrics cannot tell whether a Lately Summary is accurate or whether the reranker removed the right candidates," the engineers write. Their answer is another language model, used as a judge to grade the first ones. The post reports no figures at all: no adoption numbers, no satisfaction results, no effect on time spent. PPC Land noted the same gap in the October announcement, which gave no countries, no weighting, no persistence rules and no user counts.

Put those four together and the reframe becomes hard to avoid. Your Algorithm is less a steering wheel handed to the user than a readable label the system now keeps on every user, which the user is allowed to annotate. The recommendation engine already had a model of you. What changed is that the model now has words on it.

And there is one more detail that makes the "control" framing awkward. On November 5, The Associated Press reported that Meta, Google and TikTok had each sued California to block a new law restricting addictive feeds for users under 16. Meta's argument, as AP and other outlets reported it, is that the law violates the First Amendment by letting the government dictate how ideas are organized and shared. In court, in other words, Meta describes its ranking as its own expressive judgment. In product announcements, the same ranking is "your" algorithm. Both can be partly true. They cannot both be the whole truth.

The algorithm did not hand over the controls. It learned to describe you in words, and let you correct the spelling.

What a creator looks like from inside the menu

So far this has been about viewers. The more consequential shift is on the other side of the screen, and it follows directly from the mechanism.

If every viewer now carries a list of named topics, then every post has to be matched to those names to benefit from them. Meta's post says added interests retrieve content through search and topic clusters, and that candidates must pass a meaning-based relevance check against the interest to earn the promotion. A post that clearly belongs to a nameable topic can ride all four retrieval routes. A post that is charming but impossible to classify gets little help from anyone's list, though it can still travel on ordinary engagement.

Instagram has made that matching visible. Since the June 10 rollout across the app, according to Instagram's own announcement, posts recommended in Feed because of one of your interests carry a small label, such as "From Running." Tap it and you can see more of that topic or change your preferences. For a creator, that label is a clue to how the system filed the post. It also means the viewer now sees the filing, and can act on it.

Lucy Hall, writing for SocialDay in October, made the brand-side version of this argument: reach will increasingly depend on whether content matches topics users have explicitly named, not only on engagement. She was careful to add that Meta has not disclosed how much weight the typed inputs carry, and that holds today. Nothing in the November post gives a number. Still, the direction is clear enough to plan around, and it lines up with what the site has written before about how the Instagram algorithm works: signals about what a post is have been gaining ground on raw like counts for some time.

There is a harder edge, too. When a viewer types a creator's name into "Not Interested In," that creator is filtered out of that person's feed directly. When a viewer excludes a topic, the embedding filter removes posts that sit close to it, and the reranker sweeps up related items. None of this shows up anywhere a creator can see. PPC Land pointed out in October that a share of an audience may now be signaling "less" through a text box the account owner never sees, and that the only symptom would be a change in reach. Anyone who has stared at an unexplained dip in their Instagram analytics now has one more possible cause, and no way to confirm it.

The quieter risk is topic drift. A food creator who posts one trending dance clip for a quick spike is not only off-brand. Under this design, that clip may get filed under a topic many of their followers have never added, while viewers who excluded that topic will have it removed before they see it. The old advice about consistency was mostly about audience trust. It now has a mechanical reason as well.

An old fight in a new interface

None of this arrives from nowhere. The argument over who decides what appears in a feed is as old as the ranked feed itself, and Instagram has been on both sides of it.

Instagram dropped its purely chronological feed for an algorithmic one in 2016, and the complaints never fully stopped. In 2022 it added two alternatives, a chronological "Following" view and a "Favorites" view, but as TechCrunch reported at the time, neither could be set as the default. The company's position, then and since, has been that its research shows people are more satisfied with ranked feeds.

Europe forced the next step. The Digital Services Act requires the largest platforms to offer at least one recommendation option that is not based on profiling, and in August 2023 Meta confirmed, as TechCrunch reported, that Facebook and Instagram users in the EU would get non-personalized feeds. How easy those options were to find and keep became a court case. On October 2, 2025, a Dutch court ruled in a suit brought by the digital rights group Bits of Freedom that Meta had not complied, finding among other things that the apps reverted users to the personalized feed each time they closed the app. The Record and the DSA Observatory both reported the ruling. Meta said it would appeal.

Meanwhile, the industry moved toward a softer idea: not an off switch, but a dial. Instagram added a way to reset suggested content in late 2024. TikTok rolled out a "Manage Topics" setting, which TechCrunch described as a dozen or so generic categories. Threads launched "Dear Algo" in February 2026, in the US, the UK, Australia and New Zealand according to Meta's announcement, letting people post a request that adjusts their feed for three days. Instagram's Your Algorithm arrived for Reels in the US on December 10, 2025, reached Explore globally in English on April 15, 2026, and spread to Feed and the rest of the app on June 10, according to Instagram's own timeline. Facebook followed on October 13.

Seen in that sequence, the November engineering post is the end of a long design arc. Every step offered people more ways to adjust the ranked feed, and none of them made the ranked feed optional by default. The site's earlier look at everything that changed in the Instagram algorithm this year covered the creator-side rules. This is the viewer-side half of the same story.

Regulators want an off switch, Meta offers a dial

The timing of the explainer makes more sense against the legal calendar around it, because nearly every government pushing on Meta right now is asking for something Your Algorithm is not.

In the United States, Meta agreed in August to settle claims from a coalition of state attorneys general over teen social media use. The New York attorney general's office put the value at up to $17.1 billion, while AP has described it as up to $18 billion. Among the design changes reported by the New York attorney general and by law firm summaries is a chronological, non-personalized feed option for teens, drawn from friends and accounts they follow, with parents able to make it the default. It is due within months of the deal taking effect.

In California, the new law that Meta, Google and TikTok sued over on November 5 would bar addictive feeds, endless scrolling and autoplay for users under 16, according to AP. Governor Gavin Newsom signed it in September and it is set to take effect on January 1, 2027, unless a court blocks it. Attorney General Rob Bonta's office said it would defend the law.

In Australia, the government released a draft "My Feed, My Way" law on September 8 that would require platforms to give users over 16 a choice between personalized recommendations and posts from people they choose to follow, with penalties of up to AU$109.2 million, according to reports carried by several US public radio stations and the Australian trade title Mediaweek. Communications Minister Anika Wells said officials were still taking advice on whether algorithmic feeds should be opt-in or opt-out.

And in the EU, the European Commission issued preliminary findings on July 10 that the design of Instagram and Facebook breaches the Digital Services Act, naming infinite scroll, autoplay, push notifications and highly personalized recommender systems, as Al Jazeera and The Next Web reported. Meta disputes the findings, and they are not a final decision.

Look at what each of those asks for. A default. An opt-in. A feed without personalization. Your Algorithm is a different answer to the same pressure: keep personalization on for everyone, and make it explainable and adjustable instead. Social Media Today's Andrew Hutchinson made this point directly, suggesting Meta hopes such tools ease calls for a reverse chronological feed. That is an inference about motive, and Meta has not said it. But the product and the lawsuits are moving at the same time, in opposite directions, and anyone planning a content strategy should notice both.

For creators, the follower relationship is the part that survives in every one of those scenarios. A teen in a chronological feed, an Australian who opts out, a European on a non-profiled view: all of them see accounts they chose to follow. Building a base of real followers who actually want your posts was always the honest way to grow. On current trends, it is also the only reach that every regulator agrees should exist.

The numbers, and what they can and cannot tell you

The case for ranked feeds, from Meta's side, is built on time spent. On the company's second-quarter 2025 earnings call, Mark Zuckerberg said improvements to its recommendation systems had led to a 5% increase in time spent on Facebook and 6% on Instagram in that quarter alone. On the first-quarter 2026 call, Meta said ranking improvements lifted time spent on Instagram Reels by about 10%, and CFO Susan Li said same-day posts now made up more than 30% of recommended Reels, more than double the level a year earlier. Summaries of both calls agree on those figures.

The best outside evidence on the alternative comes from a large experiment that Meta itself enabled. In a 2023 paper in Science, a team led by Andrew Guess studied consenting Facebook and Instagram users during the 2020 US election, moving some of them to a reverse chronological feed for three months. The chronological group spent substantially less time on the platforms and was less active. Their exposure to content from untrustworthy sources rose. And their political attitudes, including polarization, did not change in any measurable way over the study period.

Those two data points explain the business logic well. Ranking keeps people around, and taking it away did not deliver the civic benefit critics hoped for, at least over three months in one election. Neither of them tells us anything about Your Algorithm, because Meta has never published a single number about it. We do not know how many people have opened the page, how many have edited it, how long an edit actually shifts the feed, or whether users who edit spend more or less time in the app. Hutchinson wrote back in June that he expected most people would never use the setting. That is a reasonable guess. It is still a guess.

For creators reading the Q1 figure, the same-day number may matter more than the time-spent one. If fresh posts make up more than 30% of recommended Reels, the system is getting faster at indexing new posts and deciding who should see them. Meta tied the gain to faster indexing; it seems likely that speed leans on classification, which is exactly the step Your Algorithm now exposes to viewers. The earlier reporting here on why saves have been beating likes for Reels reach fits the same pattern: the system keeps reaching for signals that say what a post is worth to a specific person, not just how many people tapped it. Anyone refining a Reels strategy should read the speed figure and the classification step together.

What we know, and what we don't

What is established, from Meta's own posts: Your Algorithm runs on Facebook with language models that build topic profiles, validate edits and rerank results. New interests get a temporary boost and a daily cap. "Not Interested In" entries remove close matches, and naming a creator filters that creator out. Quality requests such as less AI-generated content are handled by separate demotion signals. On Instagram, the feature has covered Feed, Reels and Explore globally in English since June 10, with labels like "From Running" on recommended posts.

What is not established: whether Instagram's version runs on the same machinery. The engineering post does not mention Instagram at all, and while the two products share a name, a design and a parent company, we have not seen Meta say they are identical underneath. Nor do we know how long the new-interest boost lasts, how large the daily cap is, how much a typed preference weighs against behavior, or whether typed preferences feed ad systems. PPC Land raised that last question in October, and Meta's posts still do not answer it.

And there is a weaker claim this piece leans on, so it is worth stating plainly: the idea that topic clarity now helps reach is an inference from the published design, not a measured result. Meta describes retrieval by topic and a relevance check against interests. It has not published any data showing that clearly classifiable posts reach more people because of this feature. The inference is sound. It is not proven.

A working plan for creators who grow without shortcuts

None of this calls for a new bag of tricks. It calls for doing a few honest things more deliberately, because the system has become more literal about what a post is.

1. Name your topic in words a viewer would type. If a stranger had to add your niche to their list, what would they write? "Budget meal prep" and "beginner bouldering" are topics. "Vibes" is not. That phrase belongs in your bio, your captions and, where it fits, the spoken opening of your Reels.
2. Write captions that classify, not just tease. Meta says interests retrieve content partly through search and a meaning-based relevance check. A caption that plainly says what the post is about gives both something to work with. The site's guide on how to write Instagram captions covers the craft; the new part is treating the first line as a label as well as a hook.
3. Keep hashtags few and literal. Hashtags matter less for reach than many older guides claimed, but a small set that names the actual subject helps a classifier as much as a person. The thinking in our hashtag strategy guide still applies: relevance over volume.
4. Treat off-topic viral bait as a cost. A post outside your topic may get filed where your audience has not asked to be, and removed for viewers who excluded it. If you want to branch out, branch into adjacent topics your followers plausibly already have on their lists.
5. Read your recommendation labels. When your posts surface in other people's feeds with a "From" label, that is Instagram telling you how it filed you. If the label surprises you, your packaging is sending a different signal from the one you intended.
6. Plan in topic runs, not one-offs. Because each interest takes turns and is capped daily, a steady stream within one topic gives the system repeated chances to match you to the right lists. A content calendar built around two or three named topics does this naturally.
7. Invest in the follow, not just the view. Every regulatory path above keeps the following feed alive. Recommendation reach can be adjusted by a text box you never see. A follower who chose you is the one audience every version of the feed still shows you to.

Getting onto the menu

The easy reading of Meta's November post is that the algorithm is finally opening up, and in a narrow sense it is. People can see a description of themselves, argue with it, and watch the feed shift a little. That is more than most of them had a year ago, and it would be churlish to pretend otherwise.

But the menu was always the kitchen's idea. Meta prints it, translates the handwriting, rations the portions and still watches what gets eaten. The people with the most to gain or lose from that arrangement are not the diners adjusting their orders. They are the cooks trying to get a dish onto the page at all.

For anyone building an audience on Instagram, the practical shift in stance is this: stop thinking of the algorithm as a judge of how engaging you are, and start thinking of it as a librarian deciding which shelf you belong on. Engagement still matters. But a post the librarian cannot shelve has a harder time reaching the reader who asked for exactly that shelf. The creators who do well in this version of Instagram will be the ones whose work is easy to name, honest about what it is, and good enough that the people who find it decide to follow. That is not a trick. It is roughly what the best ones were doing already, and the system has just started writing it down.

FAQ

What is Your Algorithm on Instagram?

It is a page that shows the topics Instagram thinks you are interested in, along with a one-line summary written by AI. You can remove topics, add new ones by typing, and tell the app what you want less of. It launched for Reels in the US in December 2025 and since June 10, 2026 has applied across Feed, Reels and Explore, globally in English. You can reach it from the icon at the top right of Reels or through Settings and activity, then Content preferences.

Does adding a topic change my feed permanently?

Not on its own, according to Meta's description of the Facebook version. A newly added interest gets extra weight for a limited period, and each interest is capped so it cannot take over the feed. After that, how much you see depends on how you actually interact with that content. Meta has not said how long the boost lasts.

Can viewers block a specific creator through Your Algorithm?

On Facebook, yes. Meta says that if a "Not Interested In" request names a creator, that creator's content is filtered out directly. The creator is not notified and has no way to see that it happened. Meta has not published the same detail for Instagram, though the two features share a design.

Is this the same as a chronological feed?

No. Your Algorithm adjusts a personalized, ranked feed. A chronological or non-personalized feed shows posts from accounts you follow in time order, without profiling. Meta offers non-personalized options in the EU under the Digital Services Act, has agreed to a chronological option for teens in its US settlement with state attorneys general, and faces proposals for an algorithm opt-out in Australia.