There is one number in productivity advice that everybody knows: after an interruption, it takes twenty-three minutes and fifteen seconds to get back to what you were doing.
It is usually attributed to a 2008 experiment on interrupted work at the University of California, Irvine. That experiment exists, it is a good study, and the figure is not in it. What is in it is a more interesting result, and one that inverts the advice built on top of the number.
Interrupted work was finished faster
Forty-eight people spent about ninety minutes in a simulated office. They played a human resources manager back from holiday, with twelve emails to answer in each condition, instructed to reply as quickly, correctly and politely as they could.
One condition had no interruptions. In the other two, a “supervisor” broke in by phone or instant message — in one case about the task at hand, in the other about something unrelated.
The interrupted conditions were not slower. They were faster.
Faster, worse
People completed interrupted work more quickly than uninterrupted work — and reported significantly more stress (p < 0.001), frustration (p < 0.007), time pressure (p < 0.01) and effort (p < 0.001). The emails were also shorter: they compensated for lost time by writing less.
Politeness did not measurably suffer. Errors did not measurably rise. The output held up, and the cost came out of the person producing it.
That is a different claim from “interruptions cost you twenty-three minutes.” It says the bill for a fragmented day is not paid in minutes at all. It is paid in how you feel at the end of it, and in the quiet compression of the work itself.
On that number
The 23-minutes-15-seconds figure does not appear in the 2008 experiment described here — we checked the paper. The observational study from the same research group, published in 2005, examined how work fragments across a real working day and is a different design answering a different question.
The point is not that the figure is invented. It is that a precise-sounding number gets attached to a paper that does not contain it, and the attachment then licenses advice — batch your notifications, protect twenty-five-minute blocks — that the actual experiment does not support and in one respect contradicts.
What does get left behind is attention, not time
The better-supported mechanism has a name: attention residue.
In a series of experiments published in 2009, Sophie Leroy had people switch from one task to another and measured performance on the second. Those who left the first task unfinished performed worse on the second than those who had completed it. Part of their attention stayed behind.
The residue was also worse for people who expected to have less time to resume the first task — the anticipation of a squeeze kept the abandoned task alive in mind.
So the cost of switching is not a fixed recovery period that ticks away and then ends. It is a fraction of your attention still allocated somewhere else, and its size depends on how the previous thing was left rather than on how long ago it was.
The notification you ignore costs about what the call costs
The most useful single finding for anyone deciding what to do about their phone is also the least intuitive.
Researchers gave people an attention-demanding task and had their phones produce notifications — sounds and vibrations — that participants did not respond to and were not permitted to check.
Performance dropped anyway. And the size of the disruption was comparable in magnitude to that seen when people actively used the phone, for a voice call or to send a text.
Which means the discipline of not answering is worth far less than it feels like it is worth. The thing that took the attention was the alert, not the reply.
Two findings from the same study that almost nobody quotes
The 2008 experiment reported two smaller results that sit awkwardly with how it is usually summarised.
The first is that the content of the interruption barely mattered. Being interrupted about the task at hand and being interrupted about something unrelated both raised stress, frustration, time pressure and effort against the uninterrupted baseline. Mental workload was highest for the unrelated interruptions, but the main effect was the interruption itself, not its subject. Advice that distinguishes between “relevant” pings and “irrelevant” ones is drawing a line the data mostly does not.
The second is that people differed, systematically. Two personality measures — openness to experience, and personal need for structure — predicted how people fared. Which is a reminder that a single recommendation applied to a whole office is averaging over a real difference in how much a fragmented day costs the person living it.
What this changes about the standard advice
Most digital-distraction advice is built on a time-recovery model: interruptions cost minutes, therefore reduce interruptions and reclaim the minutes. The evidence supports a different model with different actions.
- Silence the notification, do not resolve to ignore it. An unanswered alert impaired performance about as much as an answered one. Willpower is being applied to the wrong half of the problem; the setting is the intervention.
- Finish or park, but do not simply leave. Attention residue was worst for tasks left unfinished. Two minutes writing down exactly where you were and what comes next is aimed at the mechanism the research identifies.
- Watch for compression rather than delay. Interrupted work came in faster and shorter. If a fragmented day feels productive, that may be exactly the finding: the same output, produced under more pressure, with less in it.
- Treat stress as the outcome variable. Speed, errors and politeness held up under interruption; stress, frustration, time pressure and effort did not. Judging your day by what got done will systematically miss the cost.
- Do not budget for a fixed recovery time. There is no evidence for a twenty-three-minute tax that must elapse. What there is evidence for is a residue whose size depends on how the last thing was left.
That is a smaller and more specific set of moves than a digital detox, and every one of them is aimed at something that has actually been measured.
Try this
For one week, change nothing about how much you are interrupted. Change only what happens at the moment of interruption.
Before you switch away, write one line: the next concrete action on the thing you are leaving. Not a summary — the next action.
You are not trying to be interrupted less. You are trying to leave the task in a state that does not require part of your attention to keep holding it, which is the specific cost the research describes.
The parts of this that are genuinely hard
The 2008 experiment is 48 people answering simulated emails for ninety minutes in a laboratory. It is a clean design and a small, artificial one, and a task that can be compensated for by writing less is not every task. Work that cannot be compressed — code, care, anything with an irreducible sequence — may well show the time cost the email task did not.
The interruptions were also externally imposed and predictable in kind. Most modern distraction is self-initiated: nobody rings to ask about the internship, you reach for the phone. The experiments do not model the part where you interrupt yourself, which is arguably the larger problem.
And there is an uncomfortable implication in the speed finding. If interruptions make people work faster with no measurable loss of quality, an employer reading this paper and an employee reading it will reach different conclusions from the same result. The paper measures both the output and the cost; only one of them shows up in a performance review.
The bottom line
The famous number is not in the experiment it is credited to, and the experiment found something more useful. Interrupted work got done faster, shorter and at no measurable cost to quality — and significantly more stress, frustration, time pressure and effort.
What does the damage is a notification you do not even answer, and a task you walked away from without deciding where you were. Neither of those is fixed by protecting a block of minutes. Both are fixed by a switch and a sentence.
Sources (4)
Every factual claim in this article comes from one of these sources.
Keep exploring
Continue reading
Deeper into what you just read.
Explore more
Adjacent ground, if this was useful.
One thing worth knowing
A surprising finding. The evidence behind it. And what it means for your life. BetterYou+ sends one useful idea at a time—nothing to keep up with.