Free U.S. shipping on orders $45+ · Subscribe & save
Attention Span Test: Five Timed Minutes
Nearly every page that answers this question hands you a questionnaire about how often you lose your keys. This one puts a clock on you instead. Five minutes at a deliberately boring task, and the numbers come from what you did, not from what you said about yourself.
The five-minute task
Single digits appear one at a time, about one every 1.2 seconds. Respond to every digit by pressing the space bar or tapping the button. Do not respond when the digit is a 3. It is deliberately dull. Dull is the point: the interesting number is what happens to your accuracy in minute five compared with minute one.
If you catch yourself thinking about something else, press D or the drifted button. That is not cheating and it does not lower a score. It is one of the two things being measured. The task is visual and runs on a fixed clock, so it cannot be completed with a screen reader; the explanation underneath stands on its own without it.
Example output
Made-up numbers, shown so you can see what comes back.4 self-reported drifts and 7 missed digits, which merge into 9 separate moments off task. Average response time 372 ms.
How it drifted across the five minutes
Accuracy usually slides and response times usually get faster, not slower, as people stop checking each digit and start pressing on rhythm. That combination is the vigilance decrement, and it shows up in almost everyone.
| Minute | Responses made | Missed | Pressed on a 3 | Mean response |
|---|---|---|---|---|
| 1 | 100% | 0 | 1/6 | 398 ms |
| 2 | 98% | 1 | 2/5 | 381 ms |
| 3 | 98% | 1 | 2/6 | 368 ms |
| 4 | 96% | 2 | 3/6 | 357 ms |
| 5 | 93% | 3 | 3/6 | 355 ms |
Your last seven runs
Full five-minute runs are kept in this browser, on this device, up to seven of them. They are never sent anywhere and we cannot see them. Clearing your browser data removes them. One run is a single reading on a single day: the comparison across a week is the only part worth looking at, and even that moves with sleep, time of day, caffeine, illness and how many times you have done the task before.
Nothing logged yet. Finish a full five-minute run and it will appear here.
What the five minutes measured
The task is a continuous performance task, built on the Sustained Attention to Response Task published by Robertson, Manly, Andrade, Baddeley and Yiend in Neuropsychologia in 1997. Their version showed 225 digits over 4.3 minutes, each for 250 milliseconds with a 900 millisecond mask behind it, one starting every 1,150 milliseconds, the digit 3 the withhold target 25 times. Their healthy sample was 75 adults aged 18 to 65. This page keeps the timing and runs five minutes, about 260 digits and 29 targets.
Three numbers come out of it. The first is the median run of seconds before something interrupted you: a digit that went past without a response, or a moment you flagged by pressing D. Misses and drifts within three seconds of each other count once, because a lapse spanning four digits is one lapse. The second is your hit rate on the roughly 230 digits that wanted a press. The third is how often you pressed on the target anyway.
That third number is what the task was built around. Responding to eight digits in nine turns pressing into a rhythm, and the rhythm keeps running after you have stopped reading, so the errors come from automatic responding rather than from not trying. In Robertson’s 25-person retest group, healthy adults pressed on the target 4.56 times out of 25 on average, standard deviation 4.88. A spread nearly as wide as the average is why this page prints no comparison score.
Whether it really falls off across five minutes
Sustained attention does decay with time on task in the general case. Mackworth described the effect in radar watchkeepers in 1948, and See, Howe, Warm and Dember pooled 42 studies and 138 experimental conditions in a 1995 Psychological Bulletin meta-analysis, finding the decline in sensitivity was large on average, roughly 0.7 in standardised effect size.
Be careful reading it off your own table, though. Robertson’s group looked for a time-on-task effect inside the SART and did not find a significant one, and said why: the task has too few targets for that analysis. Five minutes gives about 29, so a worse minute five is one noisy comparison, not a demonstration. The response time row is the one to read. Getting faster while getting less accurate is the rhythm running the task instead of you.
About the 47 seconds
That figure is from Mark, Iqbal, Czerwinski, Johns and Sano, published at CHI in 2016. Forty people at one large US technology company were logged for about twelve working days by software recording which window was in the foreground. Average time on a screen before switching was 47.0 seconds; the median was 40.2. Two things get dropped in the retelling. The population was knowledge workers at their desks at one employer, not a general population and not a clinical one. And what was measured was screen switching, which the authors themselves call only a reasonable proxy for attention duration during computer work. It is not a threshold anyone passes or fails.
Your number here is a different quantity: seconds until a lapse on one dull task with nothing else on screen to switch to. A smaller number does not mean you are worse than a knowledge worker in 2016.
The goldfish never had a nine-second attention span
The claim that human attention fell to eight seconds, below a goldfish, traces to a 2015 report from Microsoft Canada’s consumer insights team, which surveyed 2,000 Canadians and recorded brain activity in 112. The eight-second figure was not their finding. It appeared in their report citing a site called Statistic Brain, whose own listed sources had no record of any such research when the BBC’s Simon Maybin went looking in March 2017. Felicity Huntingford, who has studied fish behaviour for close to fifty years, told him goldfish can perform all the kinds of learning described for mammals and birds. Gemma Briggs, in the same piece, called the idea of an average attention span pretty meaningless, because it is task-dependent.
A newer version claims 7.2 seconds from a 2026 Microsoft Research study. There is no such study. Microsoft Research’s own publication index has nothing like it, and the claim appears only on content farms, one of which cannot keep the number consistent between its headline and its body. A statistic about attention with no authors, no sample size and no journal is usually the whole story.
What this page cannot tell you
It is not a diagnostic and it screens for nothing. It says nothing about ADHD, or any other condition, in either direction. Robertson’s group called their task a laboratory paradigm and went no further than suggesting it might be a useful addition to clinical assessment: no diagnostic cutoffs, no sensitivity or specificity figures, no reference tables. Nor is there a score here we would call typical, which is why none is printed. A browser has none of the controls a lab has, and refresh rate, keyboard latency, time of day, sleep, practice and how much you cared all move the result more than what you were trying to measure.
The seven-day log has one reliability number behind it. Robertson retested 25 healthy adults a week apart and their commission errors correlated at 0.76, which the authors called stable. That still leaves around 40 percent of the variation unshared between two sittings, so expect your own number to wander without anything meaningful having changed. Pair a run with a note about the day it came from, and do not draw a trend line through seven points on a noisy measure.
If your attention or your energy has felt wrong for more than two weeks, that is worth raising with a clinician rather than with a browser tab.