Article

Interrupted mid-task: the mistakes an interruption leaves behind

An interruption does not make you generally careless. It makes you lose your place. In a 2014 experiment, interruptions averaging 4.4 seconds tripled the rate of sequence errors, the kind where you skip a step or repeat one, while errors of every other kind stayed flat. The step you were about to take is what an interruption takes. This article covers what that research measured: why position is the thing that goes, which of your work carries its sequence somewhere an interruption can reach, and what a useful check looks like when you sit back down.

Why does an interruption make you lose your place?

Eight steps into a procedure you have run a hundred times, somebody asks you something. You answer in four seconds, turn back, and carry on from step ten. Altmann, Trafton and Hambrick built an experiment around that exact gap and published it in 2014 in the Journal of Experimental Psychology: General. Their participants worked through a task made of steps, each step with its own performance requirements, and the experimenters dropped short interruptions into it.

What came out was narrow and specific. Interruptions averaging 4.4 seconds tripled the rate of sequence errors on the trials that followed them, measured against uninterrupted trials. A sequence error is a lost position: you skip a step, or you do one you already did. Shorter interruptions cost less and still cost. At 2.8 seconds, roughly the time to perform one step of the task, the sequence-error rate doubled.

The useful half of the finding is what did not move. Nonsequence errors showed no interruption effect at all. Those are the ordinary slips, a wrong key or a misread value, the mistakes that happen whether or not anybody spoke to you. Being interrupted did not raise them. It raised only the errors about position, which means an interruption is not a general tax on your accuracy.

Your place is the thing that goes.

The authors explain it through attention. The step you are on stays available because your attention is holding it there, and a momentary interruption moves that attention and cuts off the flow. Response latencies shifted too, by less than the sequence errors did. So the symptom you can feel, slowness, is the smaller half; the symptom you cannot feel is resuming at the wrong step and reading it as progress.

Chen and colleagues put a similar question to a brain-measurement setup, published in 2024 in Human Factors. Thirty-four people worked a spatial 2-back memory task under three conditions: interrupted by simple arithmetic, suspended by a fixation cross with nothing to do, and left alone entirely. Performance fell on the trials where people resumed after an interruption.

Those measures also separated an interruption from a plain pause. P3 amplitude and alpha power ran higher after interruptions than after suspensions, which the authors read as the cost of pushing the interrupting task back out of the way. An idle moment and a question you have to answer are not the same event for the work waiting underneath.

Interruptions stacked, too. Cumulative interruptions damaged accuracy more than single ones did, so the fourth interruption in an hour is not priced like the first. Read the setting before you carry any of this to your own desk. Both experiments ran short laboratory tasks with small samples, and the 2024 one measured 34 people on a memory test rather than anybody doing a job.

What travels is the shape of the cost, not a figure you can apply to your afternoon.

Which of your work goes wrong after an interruption?

Work with steps in it, where the order carries meaning and nothing in front of you records which step you finished. That is the profile the 2014 task was built to isolate, and it describes more of an ordinary week than it sounds like it should.

Onboarding a new hire through a sequence of accounts and approvals. Running a month-end close where step six assumes step five. Applying one set of edits across twelve sections of a document. Reconciling two lists by working down them together. Following a deployment or handover checklist you know well enough that you stopped ticking it. In every one of those, the order is the content, and your only copy of your position in it is the one you are holding.

Some work keeps your place for you, and it is worth knowing which of yours does. A numbered list you actually tick, a spreadsheet where the last filled row is visible, a log that writes a line per step: there the task records where you stopped, so a lost position gets caught before it becomes an error. Exposure is highest where the sequence lives only in your head.

The 2024 experiment points at the same distinction from the other side. What it disrupted was a working-memory task, holding items and matching them against what came earlier, which is precisely work that runs on something you are carrying rather than something you can look at.

If the record is on screen, an interruption costs you less.

Because interruptions accumulate, the day is a condition of the work as much as the task is. A morning with one interruption and a morning with six are not the same working environment for the same document. In the 2024 experiment the damage grew as the interruptions accumulated. Where you get any say over when the step-heavy work happens, that is usually the lever actually within reach, rather than the interruptions themselves.

How do you check work you were interrupted in the middle of?

Not by rereading all of it. The error you are looking for has a known shape, so the check can be narrow: did you skip a step, and did you do one twice.

So check the place, not the work. Find the last step you completed, confirm the one after it has not already been done, then continue. On a task whose steps leave marks, that is a two-second look at the last thing that changed. On a task whose steps leave none, it means reconstructing the position before you add to it, which is slower and is exactly when people skip the check and continue on a guess.

The obvious alternative is to stop the interruptions instead, and the evidence there is thinner than you would want. Zhang and colleagues reviewed 14 studies in 2025 in Worldviews on Evidence-Based Nursing, covering interventions meant to reduce interruptions while nurses administer medication. That is a step-ordered task with a real error count attached to it, which is what makes the review worth reading here.

Preventive measures such as no-interruption vests and awareness campaigns may reduce interruptions from outside sources. Interruptions between nurses stayed common. The effect on medication errors came out inconclusive, and the studies varied too much for the reviewers to pool them.

Read that honestly in both directions. An inconclusive error effect is not the same as no effect, a hospital ward is not your desk, and 14 studies that cannot be pooled are a thin base for anything.

So the honest split is this. Whether keeping interruptions away reduces errors is unsettled; what an interruption does to your place in a sequence is measured. No study here tested the check itself, so take it as a reading of where the measured effect sits, and as the half you control regardless of who walks up to you.

Blocking still earns its place on the work you have already identified as step-heavy. Putting that work inside a Deep Work block or a Pomodoro Technique round concentrates it into a stretch where fewer interruptions land, which is a bet on the cumulative finding rather than on the vests. Neither method was tested in any of the three studies cited here, so treat that as the mechanism applied, not as a measured result.

The other half of the problem is the interruptions you cause yourself, which this site's article on self-interruption at work takes separately, and the check above does not care which kind you had. A step you skipped because a colleague asked you something and a step you skipped because you went to look at one message are the same error with different origins.

How does your working style change the answer?

None of the three studies sorted anybody by working style, so nothing in this section is a research finding. Architect, Sprinter, Visionary and Improviser describe how somebody arranges their work, not how their memory behaves after a question. What follows is this site reading the evidence against each style's ordinary week.

If you already protect long blocks, the Architect's habit, your risk is concentrated rather than absent. A two-hour block is where the step-heavy work goes, so an interruption that gets through lands on the work with the most order in it. The cheap fix is a closing pass: before you leave the block, walk the sequence once and confirm no step got done twice. You built the block to protect the work; the pass is what protects it from the one interruption the block did not stop.

Burst workers, closest to the Sprinter, get a checkpoint for free and mostly waste it. The boundary between rounds is already a stop, which makes it the natural place to verify a position instead of resuming on momentum. A 25-minute round ends often enough that the sequence you have to reconstruct is short.

The Visionary carries the most sequence in their head and writes the least of it down, which is the worst combination this evidence describes. Thinking a problem through does not produce ticks or filled rows, so there is nothing on screen to catch a skipped step, and the work that suffers never looks like a task that could have gone wrong. Leaving a visible trace of where you are costs a line of text and removes the whole problem.

If your week never repeats, the Improviser's situation, no arrangement of your day will hold and a rule about the moment of return will. One line does it: before adding anything, find the last step that finished. That rule survives a home day, an office day and a client site without being rebuilt.

Whichever of those sounds like you, run it for two weeks before you judge it. Pick the one step-ordered task you do most often, check your position every time you come back to it, and leave everything else exactly as it is.

Then count what the check caught. A fortnight of finding nothing tells you your work already records its own place, which is a useful answer; a fortnight of catching repeated steps tells you where to put a numbered list.

If you are not sure which style is yours, the working style assessment takes a few minutes. And if what keeps breaking into the work turns out to be something other than people, the focus pillar sorts the rest of the causes.

Frequently asked questions

Why do I make mistakes at work when I get interrupted?

Because an interruption costs you your position in a sequence rather than your general accuracy: in a 2014 experiment, interruptions averaging 4.4 seconds tripled sequence errors while every other kind of error stayed flat. A sequence error means skipping a step or repeating one you already did. The authors read it as attention moving off the step it was holding available. That is why the mistake is usually one you do not notice: you resume at the wrong step and it feels like progress.

How long does an interruption have to be before it costs you?

Shorter than almost anyone expects. Interruptions averaging 2.8 seconds, about the time to perform one step of the task, doubled the sequence-error rate in the 2014 experiment. At 4.4 seconds the rate tripled. That was a laboratory task with steps of a known length, so read the direction rather than the seconds: a four-second answer to a colleague is already long enough to move your position, and the cost does not need the conversation to continue.

Which kinds of work go wrong after an interruption?

Work whose order carries meaning and whose position is recorded nowhere but in your head: multi-step procedures, reconciliations, a checklist you know well enough to stop ticking. Ordinary slips did not rise after interruptions in the 2014 data. A task with no sequence in it is less exposed than the interruption feels. Work that keeps its own record, a ticked list or a visible last-filled row, gives a lost position somewhere to be caught before it becomes an error.

Does blocking interruptions actually reduce mistakes?

Less clearly than you would hope. A 2025 review of 14 studies of interventions aimed at interruptions during medication administration found the effect on medication errors inconclusive, with too much variation between studies to pool them. Preventive measures such as no-interruption vests may reduce interruptions from outside sources, while interruptions between nurses stayed common. An inconclusive finding is not a negative one, and a ward is not a desk. No study here tested checking your position when you return, so this site puts it forward on the measured error rather than on a trial of the check.