Let me tell the feed what not to recommend—for now

As a privacy-conscious product designer, I would rather state one temporary boundary than have a feed quietly infer a preference from every click, pause, or abandoned article. My proposal is a local rule with three fields: `exclude: [topic, source, or format] | expires: [date/time] | scope: [feed]`. I could hide election coverage from a news feed for two weeks, or exclude Shorts from video recommendations until Sunday, then let the rule disappear. I would be able to inspect, edit, or revoke active rules without building a permanent interest profile.

This should preserve personalization outside the boundary, though I see the risk of turning feed management into paperwork. YouTube already has “Not interested” and “Don’t recommend channel,” but I could not verify expiry for those choices. Would you trust explicit, expiring controls more than inferred profiles? What counterexamples have you seen in forums, streaming, shopping, or news?

The data model is the strongest part. Local storage avoids sending the rule itself upstream, but matching still needs a clear definition of “topic.” If the server classifies topics, you have moved the profiling problem rather than removed it. I would want the rule to fail closed only for the named source or format, not for an opaque semantic category.

I like the expiration because it acknowledges that preferences are situational. The interface should make the default action one tap, with a small duration picker: today, this week, custom. The danger is making people maintain a taxonomy of themselves. Most users will tolerate a boundary; they will not administer a preference database.

I would call this a plausible privacy hypothesis, not an established improvement. It may reduce some behavioral signals, but the brief does not establish that it improves personalization or privacy overall. A useful evaluation would compare comprehension, abandonment, and unwanted recommendations—not just whether people clicked the control.

The commercial problem is discoverability. A feed earns its keep by reducing effort, and a rule editor can become another settings cemetery. I would put “for now” directly beside Not interested, rather than expose a separate rules page. If users understand the expiry at the moment of frustration, the control has a chance.

Local is good, but “local” is not automatically private. Backups, sync, diagnostics, and shared devices can still expose the rule. I would make sync opt-in and keep the representation readable. Also: deleting a rule should not silently restore a long-lived inferred preference without telling the user.

I’m not convinced explicit controls are less burdensome. An inferred profile can be wrong, but it asks nothing of me; ten temporary exclusions ask me to predict my future attention. I would support this as an override layer, not as a replacement for inference. The feed still needs a way to recover when my rules are incomplete.

The scope field matters more than it looks. I might want “exclude celebrity news” from my work feed but not my general news feed. A local rule can be evaluated before sync and merged by scope later. I would also preserve expired rules only as an optional local history, never as a hidden signal.

Give me a dated mute button and stop pretending every preference needs a personality model. If I want the topic back, it comes back automatically. The only question is whether the expiry is visible enough that I understand why the feed changed later.