Timing more than frequency decides whether notifications read as harassment

What it is

Notifications of identical frequency land as considerate or offensive depending on when they arrive: the subjective verdict "harassment" is driven more by timing than by frequency. Frequency answers "how many arrived"; timing answers "what was I doing when they did" — and users are far more sensitive to the second question.

Why it happens

Perceived interruptiveness is generated by context matching on the receiving side: what the recipient is doing and what state they are in at arrival determines whether a notification reads as "just in time" or "again." The same dinner reminder is useful before the meal and an offense mid-meeting — the content unchanged. Because the evaluation anchors on "right now" rather than "in total," timing contributes no less than frequency: low-frequency notifications that always arrive at bad moments get flagged as harassment faster than high-frequency ones that always arrive at good moments. The verdict, once formed, reinterprets history — a source judged harassing starts every future notification with a negative prior. This explains why un-blacklisting a notification category is so hard, and why users "suddenly" turn off an app's notifications entirely: not sudden, just the last bad timing tipping an accumulated account.

Studying it

Experience sampling (ESM) is the workhorse: immediate "was this one OK right now?" ratings after arrival, linked to logged timing and current activity. Independent variables include time of day, interval since last use, and task state (self-reported or device-inferred); dependent variables are appropriateness ratings and subsequent opt-out or blocking. Methodological cautions: the probe itself is a second interruption and inflates negativity; harassment judgments are within-person relative standards, so cross-user averages mask individual baselines — analyze within subjects.

Where it stops holding

A "good time" is not the same as a user-declared availability window — people have limited insight into when they would tolerate interruption, and the best moment shifts with task cycles. System inference of timing (from historical responsiveness) works but has a privacy price: the better the inference, the more it resembles surveillance, forcing a trade between effectiveness and chilling effects, with the inference's basis made visible. Urgent notifications are exempt; frequency's role does not vanish — volume remains the floor condition for harassment, and timing decides the direction of the verdict at a given volume.

Applying it

Make timing a first-class variable in notification policy: define unreachable and sensitive windows per category (late night sensitive by default), with non-urgent notifications in sensitive windows deferred and batch-delivered at the boundary. Timing preferences learned from history must be visible, correctable, and disableable. Verification: use immediate "was this OK?" sampling as the dependent variable, comparing immediate delivery against deferred batching — the shift in appropriateness rates tracks harassment risk better than open rates, which are contaminated by badge-clearing.

Related

• Same group: P3.06.1 Notifications can manufacture sessions the user never initiated · P3.06.2 Recall unrelated to user goals is attention taking · P3.06.3 Frequency judged by user value, not retention · P3.06.4 The cost of interruption lands on task resumption · P3.06.5 Badges are content-free recall signals · P3.06.6 Permission granularity determines whether users can refuse in part
• Adjacent: A5.14 Interruption timing and breakpoints · P3.06.3 Frequency budgets
• Search terms: perceived interruptiveness · notification timing · experience sampling