Meta today added a tool called Your algorithm to Facebook, letting people type the topics they want to see in Feed and Reels, and said large language models now help its ranking systems respond to those written requests within the same session.

In Short

Facebook now lets you type, in plain words, what kinds of posts and videos you want more or less of, and the app changes what it shows you almost straight away. That matters to anyone who depends on Facebook's recommendations to be seen, such as creators, publishers and brands, because the people scrolling can now push the algorithm in a direction without liking or commenting on anything. Meta did not say whether these typed requests are used for ads, which countries get the tool first, or how strongly a written request counts against everything else Facebook already knows about a person.

What Meta put out

The announcement arrived as a post on Meta's newsroom dated October 13, 2026, under the headline "Your Algorithm, Your Words: New Ways to Shape What You See on Facebook." It is filed under the Facebook and Product News categories. The post is short. It describes three changes and gives no figures at all: no user counts, no test results, no list of markets and no rollout schedule.

The first change is a dedicated space called Your algorithm. According to Meta, it is a place where people "can directly add and manage the topics you want to see in Feed and Reels." Meta describes it as "a dedicated space to tell us exactly what you're into." It sits inside the three-dot menu attached to posts.

The second change concerns the existing "Interested" and "Not interested" buttons. After a person taps either one, Facebook "may ask follow up questions for more detail on what you liked or disliked," according to the post. Those answers can be written freely. "You can easily tell us, in your own words, what you want to see more or less of, and your recommendations will update in real time," Meta wrote, calling it "a more direct, conversational way to shape your experience."

The third change is the redesign of that three-dot menu itself, which Meta says now makes the customisation tools easier to reach on every post.

The ranking claim

The more technical part of the post concerns the systems underneath. Meta says its ranking systems already infer interests "from the posts you like and comment on to what you watch (among thousands of other signals)." What changes, according to the company, is that large language models now sit in that pipeline.

"LLMs can incorporate the interests, context, and intent behind natural language, unlocking an algorithm you can actually talk to," the post states. Recommendations are described as "adapting within the same session to reflect what matters to you right now." Meta's summary of the engineering is a single sentence: "by bringing freeform text together with everything we already understand about what you enjoy, the algorithms behind your Feed and Reels keep getting more relevant and responsive over time."

That sentence is the core of the announcement for anyone who studies distribution on the platform. Freeform text becomes an input alongside behavioural signals. Meta's own description of its prior ranking as having "gotten remarkably good" at inference, and of the LLM addition as "a genuine leap forward," are the company's characterisations. The post offers no measurement to support either.

What the post leaves out

Several points that matter for publishers, advertisers and regulators are absent.

There is no statement on availability. The post uses "rolling out" for Your algorithm but names no countries, no languages and no date by which all users will have it. There is no mention of the European Union, where recommender systems on very large platforms carry specific obligations.

There is no statement on weighting. How a typed request is balanced against thousands of behavioural signals is not described. Nor is how long a request persists. That is a notable gap, because Meta's own sister product handled duration explicitly: on Threads, a "Dear Algo" post adjusts recommendations for 72 hours, a limit Meta published when that feature arrived in February 2026. The Facebook post speaks of updates "in real time" and "within the same session," and separately of systems that improve "over time." Whether a topic added to Your algorithm is permanent, decays or expires is not stated.

There is also no statement on advertising. The post discusses Feed and Reels recommendations only. It does not say whether the text people type into Your algorithm, or into the follow-up questions after "Not interested," is used to select ads, is stored, or is excluded from ad systems. The question is not abstract. Meta has already decided once that conversational input can shape ads: in October 2025 it said interactions with Meta AI would be used to personalise content and ads across its apps from December 16, 2025, with conversations on topics such as religion, health and political views excluded from ad targeting. Whether written feed preferences fall into the same pool is something today's post does not address.

An escalation of controls Meta already had

Facebook's users have been able to signal preferences for years, but mostly through buttons. The consumer feed on Meta's apps has long carried "Interested" and "Not interested" options, hidden words lists, a Following feed in chronological order, and Favorites. When Instagram tested a reset option in November 2024, those existing controls sat alongside a new ability to wipe recommendations in Explore, Reels and Feed and rebuild them from scratch.

The pattern since then has moved from blunt controls toward language. The reset option threw everything away. Threads' Dear Algo, launched in February 2026 in the United States, United Kingdom, Australia and New Zealand, accepted a written public post but limited its effect to three days. The Facebook version announced today is private, sits in a menu rather than a public post, and, according to Meta, persists as a managed list of topics. In that sense it is the most durable of the three.

What has not changed is that the button-based signals remain. The follow-up questions are attached to the old "Interested" and "Not interested" taps rather than replacing them.

Why ranking is now the story

For marketers the announcement is less about a menu item than about where Meta's ranking engineering has been heading all year.

Meta's earnings reports have tied engagement growth directly to recommendation work. In the first quarter of 2026, ranking improvements lifted time spent on Instagram Reels by 10%, Facebook video watch time rose more than 8% globally, and same-day posts made up more than 30% of recommended Reels on Instagram and Facebook, more than double the share a year earlier. Meta also doubled the length of the interaction sequences used to train Instagram's ranking models in that period. By the second quarter, the company was describing a Meta Generative Recommender that uses large language models to evaluate ad content and user preferences together, and reported that early pilots using LLMs to understand user preferences produced a 1% lift in app event conversions on Instagram.

Those second-quarter figures concerned ads. Today's post is the first time Meta has said, in a consumer announcement, that LLMs are now part of how organic Feed and Reels recommendations on Facebook respond to people. The two strands are clearly related in engineering terms. Meta has not said whether they share models or data.

Meta is not alone in the shift. In March 2026, LinkedIn published an engineering account of how it rebuilt its feed with LLMs and GPU-powered ranking, converting member and post features into text prompts, embedding both in a shared vector space and reading more than 1,000 historical interactions per member. LinkedIn's write-up was technical and specific. Meta's, by contrast, gives a single paragraph on architecture.

What it means for creators, publishers and brands

Organic reach on Facebook is decided by the same systems this announcement changes. Until now, a creator's or publisher's distribution depended on what Meta inferred from behaviour: what people stopped on, watched, liked, shared. A typed instruction is a new and explicit signal, and it is invisible from the outside. A page owner cannot see that a portion of an audience has written "less news" or "no more recipe videos" into a box. The effect would show up only as a change in reach.

That opacity sits on top of existing enforcement. In July 2025, Facebook reduced distribution and cut monetization access for creators who repeatedly reused others' content, having actioned about 500,000 accounts for spammy behaviour in the first half of that year. Those penalties were rule-based. Written preferences add a second, user-driven filter on top, one that rewards matching what people say they want rather than only what they click.

For paid media the link is less direct, but not absent. Advertisers running Advantage+ campaigns hand targeting decisions to Meta's models, which draw on what Meta understands about each person. If typed interests become part of that understanding, the quality of the signal could improve, because stated intent is often more precise than inferred intent. If they do not, Meta will be holding a body of explicit, self-declared preference data that its ad systems deliberately ignore. The post does not say which applies.

There is also the question of placement. Feed and Reels are where most of Meta's ad impressions appear. A user who tells Facebook to show fewer posts of a certain kind is reshaping the organic content around which those ads are served, even if the ads themselves are selected separately.

The regulatory backdrop

Personalised recommendation on Meta's apps is currently the subject of action in Europe and the United States, and today's change touches it directly.

On July 10, 2026, the European Commission preliminarily found Meta in breach of the Digital Services Act over the design of Instagram and Facebook, citing infinite scroll, autoplay, push notifications and personalised recommender systems. The Commission said it expects Meta to rework its recommender so that it is less engagement-oriented. A finding is not a fine; the ceiling is 6% of global annual turnover, and Meta can respond before any final decision. A control that lets users state preferences in their own words could be presented as user empowerment in that context. Equally, an algorithm that adapts "within the same session" to keep recommendations relevant is precisely the kind of responsiveness regulators have questioned.

In California, Judge Edward J. Davila found on August 5, 2026 that ranking by predicted engagement is "not an expressive judgment," denying preliminary injunctions sought by Meta, TikTok and Google against parts of Senate Bill 976. Weeks later, a consent judgment in the youth social media litigation committed Meta to offering teens a non-personalized default feed option, chronological and drawn from accounts they follow or have friended, with that option due by late December 2026 under the agreement's four-month clock.

Against that background, the timing of an announcement framed around user control is worth noting. Today's post does not mention teens, regulators or any legal obligation. It also does not say whether Your algorithm is available to teen accounts. The Meta newsroom page lists among its related items a May 12, 2026 post about supervision tools giving parents insights into their teen's algorithm, which suggests Meta already exposes some form of interest data to parents, though today's post does not connect the two.

The open questions

Three things will determine whether Your algorithm matters beyond the menu.

One is scale. A tool buried in a three-dot menu may be used by a small minority. Meta has given no figure for how many people used Instagram's reset option, Threads' Dear Algo or the existing feedback buttons, and none for this one.

Another is persistence and weight. If a typed request carries strong, lasting weight, it becomes a meaningful lever over distribution. If it is a short-lived nudge, as on Threads, its effect on creators and publishers will be marginal.

The last is data use. Written statements of interest are among the most explicit preference data a platform can collect. Whether that data stays in organic ranking, flows into ad delivery, or is retained for model training is not addressed in today's announcement, and it is the point on which privacy regulators in Europe are most likely to ask questions.

Timeline

Summary

Who: Meta, for Facebook users, with consequences for the creators, publishers and advertisers whose reach depends on Feed and Reels recommendations.

What: A new Your algorithm space in the three-dot menu where people can add and manage topics, written follow-up questions after "Interested" or "Not interested," and large language models in the ranking systems so that typed requests update recommendations within the same session.

When: Announced today, October 13, 2026. Meta describes Your algorithm as rolling out but gives no completion date.

Where: Facebook Feed and Reels. Meta named no countries or languages and did not mention the European Union.

Why: Meta says freeform text, combined with behavioural signals, makes recommendations more relevant and responsive. The change also arrives while regulators in Europe and courts in the United States are scrutinising engagement-driven recommender systems, and Meta did not say whether the typed data is used for ads.