Time on task

Attention span test that measures the drift, not a duration

Answer odd or even on one digit at a time for five minutes and the page reports what changed while you did it — your median answer time, your accuracy, your spread and how many digits you got through, split into one row per minute, with a 95% interval on the difference between your first minute and your last. It is free, the result is on screen the moment the run ends, and nothing is stored anywhere. There is no attention span in seconds to report because no such number has ever been measured: the famous eight-second figure has no primary source, and what a task like this can honestly show is the slope rather than the length.

100% free
No signup
5 minutes
One row per minute
No score to beat

A digit appears. Press F if it is odd, J if it is even, and the next one follows 250 ms after your answer. That is the whole task for 5 minutes. Nothing about it gets harder and nothing is ever withheld — the only variable in the run is how long you have been doing it, which is the only way to measure what the phrase in the search box is actually asking about.

The run, spelled out

Digit: one numeral from 1, 3, 7, 9 and 2, 4, 6, 8, white on a dark panel, no color anywhere
Answer: F for odd and J for even, or the two buttons under the panel
Pace: 250 ms of cross, then the digit, then 250 ms after your answer — you set the rest
Length: 5 minutes of scored digits, cut into 5 segments of one minute
Window: 1500 ms; past that the digit is gone and recorded as missed rather than dropped
Order: parity runs capped at 4, and no digit ever follows itself
Void: answered inside 150 ms, answered before the digit painted, or caught by this window losing the front

Parity was chosen because it is the least interesting decision that still needs looking: there is nothing to learn, no strategy to find and no way to get better at it, so anything that changes over the run is the watching rather than the task.

The one-minute option produces a single point and therefore no trace at all: there is no second minute to subtract the first from. It is here so you can find out what you are agreeing to before you agree to 5 minutes of it.

How to run the attention span test

One key for odd, one for even, and five minutes of nothing else happening.

Put a finger on F and a finger on J and leave the room alone
F is odd, J is even, and the digit is always one of 1, 3, 7, 9, 2, 4, 6 or 8. Nothing about the task changes at any point, so there is nothing to learn and no strategy to find. If you have no keyboard the two buttons under the panel do the same job, though a run that mixes tapping and typing puts a change of method inside the change the page is about to measure. Close the messaging app first: the run voids any digit that was on screen while this window was at the back, and a five-minute measurement full of voids is not a five-minute measurement.
Take the twelve practice digits if the keys are not yet automatic
Practice gives you 2,500 ms per digit instead of 1,500 and names the answer back to you after each one. None of it reaches any figure. Its real job is to get the finger mapping out of the way before the clock starts, because a first minute spent remembering which key is which produces a fast-looking improvement afterwards that has nothing to do with attention.
Read the trace rather than the total
At the end you get five rows, one per minute, each carrying the number of digits you answered, your median answer time, the spread of that minute's correct trials and the share you got right. Under them sits the last minute minus the first with a 95% interval on it, and a note saying where your own slowest tenth of trials fell. A flat trace across five minutes is the ordinary outcome and not a good score; the shape worth looking at is a widening spread with a steady median, which is what a person who has started missing occasionally looks like before their average moves at all.

Technical specifications

Run length	Five minutes of scored digits, cut into five one-minute segments by the clock rather than by trial count. A one-minute option exists to see the task and produces one segment, which is to say no trace at all
Task	Parity of a single numeral: odd for 1, 3, 7, 9 and even for 2, 4, 6, 8. Chosen because it is the least interesting decision that still requires looking — no learning curve, no strategy and no ceiling to run into
Pace	Self-paced. 250 ms of fixation cross, then the digit, then 250 ms after your answer before the next one — roughly 270 digits in five minutes at an ordinary speed. Because you set the rest, slowing down shows up as a lower trial count as well as a longer median
Response window	1,500 ms from the digit painting. Past that the digit goes and the trial is recorded as missed rather than dropped, so a lapse is visible in the counts instead of quietly shortening the run
Sequence	A parity-balanced stream shuffled with runs of one parity capped at four, and no numeral ever immediately repeating itself — a repeated digit is answered from the previous answer rather than from the screen
What voids a trial	An answer inside 150 ms of the digit painting, an answer that arrives before it paints, and any digit that was on screen while this window was behind another. A digit lost to this window going behind another is asked again later in the run; the other two are not, because a hand that went early has already seen the digit. All three are counted out of their minute rather than into it
Reported per minute	Trials answered, median answer time, standard deviation of that minute's correct trials, and share correct. Plus the run total, the outlier rule that produced it, and the count of trials it removed
Reference figure	One, and it is a warning rather than a benchmark: a monitoring task takes roughly 30 minutes before detection measurably falls (Mackworth, 1948). Five minutes ends long before that, which the results panel says out loud

Frequently asked questions

Is the human attention span really eight seconds, shorter than a goldfish?

No, and there is no study behind the claim at all. Following it backwards leads to a 2015 consumer report published by Microsoft Canada, which attributed the figure to a statistics aggregator whose own sourcing cannot be verified, and the goldfish half has no traceable origin of any kind. A 2017 BBC investigation by Simon Maybin went looking for the underlying research and could not find any. The deeper problem is that the quantity does not exist to be measured: how long someone stays on something depends on the task, the stakes and whether they care, which is why a person who cannot sit through a two-minute video also watches a three-hour film.

Five minutes seems short. Why not run it for half an hour?

Because almost nobody finishes a thirty-minute browser test, and a measurement half the visitors abandon is worse than a shorter one they complete. It is worth being explicit about what the length costs: the classic vigilance decrement, where detection measurably falls off, takes around half an hour of watch-keeping to appear. What five minutes can show is the earliest part of that curve — the spread of your answer times widening, and long trials starting to cluster near the end — which moves before any average does.

My last minute was slower than my first. Does that mean my attention is poor?

It means this run slowed down, and the 95% interval printed beside it tells you whether the difference is bigger than the ordinary wobble between two minutes of the same task. Most runs come back with an interval that spans zero, which is the honest answer for five minutes of data. A drift of thirty or forty milliseconds over five minutes is also just as likely to be a hand settling out of an over-eager start as attention giving way, and this task cannot separate those two.

Why does the page care about the spread of a minute rather than its average?

Because attention fails intermittently rather than uniformly, and an average hides exactly that. Someone whose median holds at 480 ms while their standard deviation grows from 90 ms to 160 ms has not slowed down at all — they have added a handful of very long trials, which is what a lapse looks like in a stream of otherwise ordinary responses. That is also why the page counts where your own slowest tenth of trials landed: an even scatter across the five minutes means those long trials were noise, and a pile-up in the last two means something else.

Would a harder task measure attention better?

A harder task measures the difficulty as well, which is the opposite of what is wanted here. If the digits had to be added, or held in mind, or compared with something, then a change over five minutes could be a change in arithmetic or in memory rather than in staying with the job. Parity is deliberately at the floor of what still requires looking at the screen: there is no skill in it to improve and no load to run out of, so whatever moves across the run has only one place left to come from. It is also why this page is not a working-memory test and does not pretend to be one.

Can I compare my median here with a friend's, or with a figure I saw elsewhere?

With a friend on the same machine, cautiously; with anything else, no. The absolute level of an answer time is mostly a fact about the display, the keyboard and the browser — the same person routinely differs by fifty milliseconds between two laptops — so the level travels badly. What travels is the shape: your own first minute against your own last, measured on one screen in one sitting, with everything constant except how long you had been at it. That is why this page reports a difference with an interval and no rank of any kind.

Does answering with the on-screen buttons change the result?

Yes, by tens of milliseconds a trial, and the results panel says how many of your answers arrived that way. Aiming a pointer at one of two buttons costs a movement that a resting finger never makes, so a tapped run sits uniformly higher. That constant offset largely cancels out of the minute-to-minute comparison, which is the figure this page leads with — but switching between tapping and typing partway through does not cancel, because the switch lands inside the drift and is indistinguishable from it.

Where the eight-second figure came from, and what can be measured instead

The claim that human attention has collapsed to eight seconds is the most widely repeated statistic in this subject and one of the least supported. It appears in conference talks, marketing decks and school assemblies, always with the goldfish attached, and it survives because it is repeated rather than checked. When it is checked, the trail runs backwards from a 2015 consumer report published by Microsoft Canada to a statistics aggregation site, and stops there: the aggregator’s own citation cannot be resolved to any study, and nobody has produced a paper measuring the quantity in question. The goldfish comparison has no origin at all.

Why this page prints no attention span in seconds

The claim that the average human attention span fell to eight seconds, below a goldfish, has no primary source. It has been traced repeatedly to a citation chain that terminates in nothing, and no study measuring any such quantity has been produced. Attention span is not a single duration in any case: it depends on the task, the stakes and the person's interest.

What does exist is a large and old literature on what happens to performance while a person stays on one job, and its central finding is that the failure is gradual, partial and mostly invisible in an average. The reference point is a wartime study of radar operators watching a pointer for a rare irregular jump: Mackworth (1948), The breakdown of vigilance during prolonged visual search, Quarterly Journal of Experimental Psychology ran the watch for two hours and found detection falling off measurably after roughly the first half hour. Three things about that finding matter for a five-minute browser task. The timing depends on how often the signal arrives and how obvious it is — rare and faint degrades a watcher faster than frequent and clear. The decline is in detection, not in the ability to do the task when engaged. And half an hour is six times longer than this page runs, which is why the results panel says plainly that a flat trace here is the expected outcome rather than a good result. What five minutes reaches is the part that moves first: the spread of the response times, and where the slowest trials sit in the run.

The design choices follow from that. The task is parity because a task with any difficulty in it lets a change over time be a change in the difficulty instead; the run is self-paced because a fixed beat would hide slowing inside the gaps it imposes; the segments are cut on the clock rather than on trial count because a minute is the unit the question is asked in. Three neighboring pages take the other routes through the same territory: the continuous performance test runs ten minutes with a target that almost never comes, which is the closest a browser gets to Mackworth’s watch; the concentration test puts a clock on each row of a dense cancellation sheet and reports the swing between rows; and the selective attention test asks what competing distractors cost rather than what time does. If what brought you here is a general curiosity about where your abilities sit, the practice IQ test and the numeracy test answer different questions entirely, and the Posner cueing task measures how attention moves rather than how long it lasts.

A reaction time here is the interval between the frame that painted the stimulus and the timestamp the browser attached to your key, both read from the same monotonic clock. What neither can see is the display pipeline behind it, so on a 60 Hz screen roughly 16 ms of every figure below is the machine rather than you. That is the timing floor: two numbers closer together than that are the same number, and this page reports no precision it cannot support.

This is a measurement exercise, not a clinical assessment. It reports what you did on this page against a stated reference and nothing more — it cannot establish ADHD, an attention disorder or a concentration problem of any kind. Only a qualified professional, working with more than a browser, can make that judgment.

Where your five minutes of keystrokes go

Every number on this page is worked out by JavaScript running in the tab you are reading it in. Your answers, your reaction times and your score are never uploaded, logged or kept — which is also why the test carries on working after you disconnect from the network, and why nothing here can be held back behind an email address.

Roughly 270 timestamps accumulate in one JavaScript array in this tab and are used for two subtractions and a handful of medians. Nothing is written to storage, so a refresh loses the run and the copy button is the only way to keep it — which is also why the copied text carries every minute rather than a single headline figure.