Negative feedback must produce an immediately visible change, or users will judge it useless and stop

What it is

“Not interested” is tapped; the row stays; it stays after refresh; the class is still there tomorrow. The only conclusion available is that the act is not wired to the world. Immediate negative-feedback effect means a reject must leave an observable consequence on this screen — the row gone, the scope echoed, the class dropped at once — or the channel is judged dead, and even rejects that would have worked later stop happening.

Scope has to be right. It also has to be visible as already in force.

Why it happens

Instrumental acts are kept alive by immediate causation: press, world changes. Recommender learning is often batched; the real weight update lands in hours or days. If the interface waits for the batch, the causal chain is broken on the user’s timescale. People will not keep paying an operating cost for an unseen future update; use of the channel collapses after the first “nothing happened.”

Immediate visibility need not mean the model has already retrained. It can be a UI-level promise: this row leaves the current list first, this class is down-weighted in this session, the back end catches up. Without that layer, however punctual the back end, the user has already left the channel. Visible change also calibrates: people understand scope from what happens in front of them; if nothing happens, a clearly written scope is still empty talk.

Studying it

The same reject, three delays: remove from the list and echo at once, change on the next page, change next session. Measure: whether the channel is used again, whether people switch to coarser means (rage-scroll, turn recs off, a one-star rant), understanding of scope. Independent variables: delay, whether an “handled” echo exists even if the list is briefly unchanged. Dependent variables: reuse, channel abandonment, judging delay as “did nothing.”

An echo is not a change. Copy only, row still there, will still be judged dead. Report “words” and “list” separately.

Where it stops holding

Moderation rejects (report) cannot be pulled from the whole network at once; what should change at once is “hidden for you” and case status, not site-wide disappearance. Offline or on a weak network, at least honour it on the local list and mark that it will sync. This entry wants immediate visibility. It does not treat emotional semantics, and it does not treat one reject wiping a class — that is scope too large, not delay.

Applying it

• After a reject, the current list must change at once in line with the chosen scope: the row gone, or the class dropped in remaining seats on this screen, plus one echo sentence.
• Back-end retraining can be late; in-session filtering must be immediate. Do not treat “added to training” as a change the user can see.
• Check: within three seconds of the press, the screen must differ in a way that matches scope. If it does not, watch whether people use the channel a second time. A collapse in second use means the channel is already dead on the user side, regardless of when the model updates.

Related

• Same group: L6.13.1 With only positive signals the system cannot tell "not interested" from "never saw it" · L6.13.2 Negative feedback must state its scope: this item or this class · L6.13.4 Users treat negative feedback as emotional expression; its semantics are imprecise · L6.13.5 Over-responding to a single negative can wipe out a whole class of content
• Nearby: L6.07 Presenting Recommendation Reasons · L6.10 Turning Personalization Off and Resetting It · L1.06 Graceful Degradation of AI Failure
• Search terms: immediate negative-feedback effect · visible reject · channel abandonment