Beyond about ten seconds, users turn to other tasks
Aliases: 10-second rule · attention wander · dialogue attention span
What it is
The dialog is still open, progress is still moving, and the person has already picked up a phone to answer a message — not out of rage, but because attention has left this dialogue. About ten seconds is the empirical ceiling for holding attention on the current human–computer conversation. Past it, people demote the wait to a background affair and go do something else. That bound is the ten-second attention ceiling.
It is not “ten seconds is the most anyone will tolerate”. People can wait longer. What changes is ownership of attention: after ten seconds the interface no longer holds the main channel, only a thin monitor of “glance back if something happens”.
Why it happens
Attention is a costly, sustained investment. In a short wait, watching progress and holding the goal are the same act. Near ten seconds the return on monitoring drops — the picture often has no new information, and the goal is still unfinished. People cut the main channel away and leave a narrow one for “call me when it changes”. Miller marked ~10 s as the bound for holding attention on a dialogue; Nielsen reused it as the third usability band of response time. The number moves with context; the jump does not: from “I am waiting on it” to “it is going on somewhere”.
Once main attention leaves, every earlier UI promise is repriced. A quiet green flash at completion will be missed; a mid-wait confirmation will find nobody home. The design object is no longer “the person watching” but “the person who will come back”.
Studying it
Give waits of 3, 10, 20 and 40 s, allow a phone or another window, and mark the first clear task switch (unlocking the phone, another tab, standing up). Do not forbid switching — forbidding yields a fake “they stared the whole time”.
Independent variables: wait duration, whether new information appears during the wait, whether other work is possible during it. Dependent variables: time of first task switch, whether the original window is still glanced at, whether return can resume.
Labs with no real “something else” overestimate ten seconds. Field work, or a genuine secondary task, sits closer to the product ceiling.
Where it stops holding
In a cinema, an exam, or a cockpit, external tasks are locked out and people can watch a progress bar for tens of seconds. The ceiling did not fail; switching was forbidden by the environment. Game loads and video buffers treated as part of the experience (there is something to watch) can keep attention inside the product, but not on the original operational dialogue. Urgent actions (a payment countdown, a safety confirm) compress the ceiling: a few seconds without progress and people intervene rather than leave. Screen-reader users may already be near ten seconds after one spoken prompt; “still in the dialogue” has to be counted in spoken turns.
Applying it
- Recast operations that may exceed ten seconds from “please watch” to “you may leave”: do not pin people to the wait with an un-dismissable modal.
- Design the wait on the assumption that main attention will leave, so completion cannot rely on an in-place colour change alone.
- Split into stages shorter than ten seconds where you can, so people can decide at a stage boundary whether to stay or go.
- How to check: make an export last 25 s and put a usable phone beside the person. Note when the phone is first picked up. If it clusters near ten seconds, the product must be designed for leaving, not for staring.