Facebook Feed and Reels now take plain-language instructions on what to show more or less

Take a Reels viewer who gets a fourth clip on a topic they have had enough of. Meta now gives them two ways to answer. They can tap "Not interested", and The Verge reports that button may also appear directly under posts ^[4]. Or they can say, in their own words, what they want less of ^[2].

These are different tools, and the pitch blurs them. Meta says LLMs are "unlocking an algorithm you can actually talk to" and that recommendations respond "in real time, adapting within the same session to reflect what matters to you right now" ^[8]. What the screen offers is a menu entry, two buttons and optional follow-up questions ^[3][5]. A button takes one tap. A sentence takes effort, and we think the follow-up questions exist because some of those sentences will be vague.

The testable part of the promise is "within the same session". A user can write an instruction, keep scrolling and see in one sitting whether the topic thins out. If it does not, the user finds that out in the same sitting.

Neither report includes usage figures from Instagram, where the control already exists, and neither says how a stated preference is weighed against what a person actually watches ^[6]. Meta's stated goal is more relevant recommendations ^[11]. Count second uses of the control and whether the unwanted topic thins. Time spent scrolling afterward measures something else.

We think the sentence box suits a user who can name the annoyance, and the buttons suit one who can only point at a post.

For a team deciding whether to build the same thing, two tests sort the decision. The first is whether the feed visibly changes inside the session. The second is whether the user can see what the system understood. A feature that passes both is a control. One that changes the feed but hides what it understood is a guess that looks like a control, and its cost arrives as tickets from people who asked for less of something and still see it. If it shows what it understood but the feed does not change, it is a survey, and if it does neither, it is a suggestion box.

On Meta's own account, Facebook's version passes the first test, and the follow-up questions are its only stated attempt at the second ^[5][8]. We would call it a control on the first axis and unconfirmed on the second until someone measures it. If you build your own, we'd ship the one-tap version first and add the free-text box only when you can show the result inside the session, because a free-text box promises more than a button does and every miss is then visible to the person who typed it.