design-callouts

A callout spends the reader's attention to redirect it. Callouts fail in two opposite ways: too quiet — an error vanishing in a five-second toast nobody saw — or too loud — a page of warning boxes readers have learned to skip. Both come from choosing the component before classifying the message. Every rule below derives from three commitments:

• Match disruption to consequence. Interruption is a cost; pay it in proportion to what happens if the message is missed. A missable confirmation earns a transient toast; a data-loss warning the user can keep working alongside earns a persistent banner; a decision the user must resolve before safely continuing earns an alert dialog. Escalating past the consequence trains users to ignore the tier; underspending loses the message.
• Never the only channel. Anything the user must see or act on needs a persistent home — the main content flow, an inline message, a notification log. Transient and hover-triggered surfaces may echo essential content that already has a persistent home, but must never be its sole source. Throughout this skill, content or an action is essential when losing it makes the task impossible, unsafe, or unrecoverable; an action is essential when no other route to the same outcome exists.
• Every callout devalues the next. Attention habituates: clinical-alert studies show 49–96% of interruptive alerts get overridden once they become routine. Treat callouts as a budget; when two compete, merge or demote one.

These rules apply identically when designing from scratch and when auditing — auditing is holding what exists against the same commitments and reporting each violation with its fix.

Classify the message first

The three branches follow different rules; misclassifying is the root error behind most callout failures. Pick by trigger:

Keep the branches clean: education surfaces must not carry errors or status (a teaching tip reporting a failure will be dismissed as marketing), and status surfaces must not carry promotion.