Feeds should show which signal they are using

As a product designer, I want recommendation feeds to label why each item appeared: explicit follow, single click, repeated viewing, or an inference from similar users. Those signals do not carry the same meaning, but interfaces often flatten them into “because you watched something.”

A concrete failure mode: I linger on one upsetting clip, then search an emotionally charged topic to understand it. If watch time and search history influence recommendations, that brief episode can start looking like a durable identity. Maybe it is, maybe it isn’t—but I should be able to see the distinction.

I’d add a “signal inspector” beside each recommendation, with switches for each signal type. Disable repeated-viewing or similar-user inference without deleting everything else. Would transparent labels genuinely restore control, or just make profiling feel more acceptable? I’d like real examples, especially where a feed misread you.

The useful part is separating attribution from explanation. A feed may have several contributing signals, and labeling one as “the reason” could imply a certainty the system does not have. I’d want ranked signal weights or at least “contributed to,” plus a way to see whether the label applies to this item or the broader model.

The per-signal switch is much better than a giant history reset. It preserves the user’s chosen personalization while creating a boundary around sensitive behavior. I’d still ask whether disabling a signal also removes previously derived inferences; otherwise the control may only stop future collection while leaving the old profile intact.

Small evidence distinction: the four labels are a sensible proposed taxonomy, not a standard interface I’d assume platforms already provide. Documented systems do use clicks, watch time, searches, follows, and similar-user comparisons, but that does not establish that any one signal caused a particular recommendation.

This sounds good until the inspector becomes a maintenance console. Most people will not tune four switches per topic, especially when the feed changes constantly. I’d make the labels visible, but pair them with one plain-language action: “Don’t use this kind of behavior for recommendations.”

I’m not convinced transparency equals control. A label can teach me exactly how I’m being profiled while leaving the platform’s underlying objectives untouched. Still, that discomfort is useful: if users reject the labels or switches, we learn the design is cosmetic rather than empowering.