Show me why this post was recommended—and let me turn off that reason

As a privacy-minded product engineer, I think every recommended post should name its specific reason: followed topic, friend interaction, watch history, or inferred interest. More importantly, that reason should have its own mute switch. I should be able to stop using watch history without deleting my whole profile or accepting a completely unpersonalized feed.

The catch is that transparency can become theater. “Because you watched…” may be only one visible input while hidden weights, eligibility rules, popularity, experiments, and inferred interests still shape the ranking. Naming a friend, group, or sensitive inference could also expose a social connection I did not expect to make visible.

For example, briefly watching medical-condition videos while researching for someone else could produce repeated health recommendations. The explanation would clarify the problem, but might itself feel uncomfortable on a shared screen. Would you actually inspect and adjust per-signal controls, or would you prefer a genuinely low-profile feed?

I want the controls, but not the fiction that one label explains the system. The UI should distinguish “contributed to eligibility” from “caused this ranking.” Otherwise users will infer a level of causality the system cannot honestly provide.

The social reason is where this gets awkward. “Because your colleague liked it” may be useful, but it also turns a quiet interaction into a visible social fact. I would default those explanations to broad categories and make the detailed version opt-in.

Per-signal switches sound excellent until the settings page becomes a cockpit. I’d put the control directly under the explanation, with a reversible “less like this” action, and hide the full taxonomy behind an advanced view.

I’d be careful about claiming this improves recommendation quality. The brief supports explainability and controllability as separate goals, but not that people routinely use granular controls or that those controls reliably produce better feeds.

An inferred-interest label can leak more than the post itself. Even if nobody else sees the setting, a compromised account, shared device, or shoulder surfer might. Privacy-preserving explanations should reveal the control path without exposing the underlying sensitive inference.

A low-profile feed is cleaner, but it may just move the profiling out of sight. I’d rather have an imperfect explanation and a way to disable inputs than pretend chronological ordering is the only honest option.

Most people probably want the feed to stop being weird, not a seminar on ranking models. Give me one tap to remove the reason, remember it, and do not make me audit a dashboard every week.