Watching a video does not always mean wanting an entire feed of similar content. You might be researching a one-off purchase, following a passing curiosity or simply wondering why a clip attracted so much attention.
On October 13, 2026, Meta announced Your algorithm for Facebook. The feature raises an interesting question: can explaining your interests make recommendations more useful than leaving the platform to infer them?
What is Facebook introducing?
Meta describes a space for adding and managing topics in Feed and Reels. It also says users may receive follow-up questions after selecting “Interested” or “Not interested,” allowing them to explain their preferences in their own words.
Our interpretation is that the useful change is greater precision. Someone who enjoys photography might want practical lighting tutorials while having little interest in camera shopping. A broad interest label does not express that distinction very well.
A written preference could potentially describe the subject, the format and the reason a person finds it useful. Whether the resulting recommendations reflect those details is the practical test.
Why use a language model for recommendations?
Meta says large language models help interpret the interests, context and intent in written feedback, alongside its existing recommendation signals.
To understand the idea, consider these two hypothetical preferences:
- “I want beginner photography tutorials.”
- “I want beginner photography tutorials using a phone, rather than reviews of expensive cameras.”
The second sentence includes a subject, a skill level, a type of equipment and an exclusion. Interpreting how those elements fit together is a language-understanding task.
This example illustrates the concept; it does not describe Meta’s internal implementation or a tested result. The announcement does not provide enough technical detail to reconstruct how an individual sentence changes the ranking of specific posts.
It also shows why AI has uses beyond generating answers. Here, language understanding supplies information to a system that selects existing content.
Where can you find Your algorithm?
Meta points to the three-dot menu on posts. Where the feature is available, look for “Your algorithm” to manage topics, or use the feedback options on a post.
Try one clear preference first. If several interests change at once, it becomes harder to judge which instruction produced a useful difference.
For example, a gardening enthusiast could ask for more practical advice about growing herbs on a small balcony. That gives a more concrete target than a general request to improve the feed.
These are suggested ways to evaluate the tool, not guarantees about the requests it will accept or how it will respond.
More influence does not mean choosing every post
AI-based ranking is not new to Facebook. In a 2023 explanation of its recommendation systems, Meta described combining predictions about how valuable content might be to a person, informed by behavior and feedback. It acknowledged that no single prediction perfectly captures whether a post is valuable.
That background helps explain the distinction between a preference and a filter. A preference tells a system what you would like it to prioritize. A strict filter would exclude every item that fails a defined condition.
The new announcement does not establish that a written preference acts as an absolute filter. Asking for fewer videos about a subject should therefore not be treated as a verified promise that the subject will disappear.
Nor does relevance establish accuracy. A feed can become better matched to someone’s interests while still containing claims that need checking. A recommendation is a selection decision, not a factual endorsement.
How could you tell whether it helps?
Meta says recommendations can adapt within the same session. That is the company’s stated behavior, not a DemystIA measurement.
A useful personal test would separate three questions:
| Question | What to look for |
|---|---|
| Did it understand the preference? | The requested subject and format appear, rather than content that merely shares a keyword. |
| Did it reduce unwanted content? | The specific type of post you wanted less of becomes less prominent. |
| Did the feed become more useful? | You find worthwhile material without having to repeatedly correct the same misunderstanding. |
One well-matched post would not establish that the change is reliable. A small personal trial can indicate usefulness, but it cannot isolate every influence on a changing feed.
For instance, request practical recipes, then check whether the recommendations actually explain preparation or simply show attractive finished dishes. Both concern cooking; only one may serve the intended purpose.
Is it available to everyone?
The announcement describes a rollout but does not give a country-by-country schedule or a supported-language list. It does not establish availability for every account.
If the option is missing, the announcement alone is insufficient to diagnose why. Avoid assuming that a screenshot from another account confirms what should be present in yours.
That distinction matters for tutorials: instructions should describe where an available feature lives, while making clear that access has not been verified for every reader.
What this means for users and creators
For users, the opportunity is to explain a preference that their viewing behavior does not express clearly. The most useful outcome would be less time dismissing unsuitable recommendations and more time finding material worth reading or watching.
For publishers and creators, our interpretation is narrower: this announcement provides no basis for promising a reach increase or a new ranking shortcut. A sensible editorial response is to make the intended audience and purpose of a post clear.
A specific tutorial should explain what it teaches and who it helps. That is useful to readers regardless of whether a recommendation system gives additional weight to their written preferences.
The development also fits a broader pattern: AI becoming part of familiar software interfaces. Our article on Firefox Smart Window and Mistral examines another example, where an assistant works with browsing context.
Facebook’s new controls are worth evaluating through a simple question: does explaining what you want produce a feed that serves you better? The answer depends on the recommendations people receive, not just the convenience of the input box.
Sources
- Meta — Your Algorithm, Your Words: New Ways to Shape What You See on Facebook, October 13, 2026. Official feature announcement.
- Meta — How AI Influences What You See on Facebook and Instagram, June 29, 2023. Historical background on ranking and predictions; not a specification of the new feature.
Information checked October 15, 2026. This article analyzes Meta’s public announcements. The preference examples and evaluation suggestions are illustrative, not results from a hands-on test.