You have almost certainly seen it: after an interruption, it supposedly takes 23 minutes and 15 seconds to get your focus back. The figure shows up in remote-work explainers, corporate productivity training decks, and roughly half the focus-timer apps on the market. Its precision — down to the second — makes it sound like the output of a controlled experiment.
There was no controlled experiment. The number appears in no published scientific paper. It comes from a 2006 press interview, it describes something quite different from what people claim, and the researcher who said it originally offered it as reassuring news.
The number came from an interview, not a paper
The source can be dated to the day. On June 8, 2006, the Gallup Business Journal published an interview with Gloria Mark, then an associate professor at the University of California, Irvine, and a specialist in workplace interruptions. Asked what happens to interrupted tasks, she answered that most are resumed the same day — 81.9 percent — and that on average work was "resumed, on average, in 23 minutes and 15 seconds, which I guess is not so long".
That closing clause deserves attention. The number the productivity industry now brandishes as an attentional catastrophe was delivered in a tone of relief. It described a reasonably fast return, not a loss.
In 2023, a developer set out to trace the number back to its original publication. He reviewed the five papers routinely cited as its source — including The Cost of Interrupted Work, the one invoked most often — and found the figure in none of them. Not 23 minutes, not 15 seconds. The most famous number in the attention literature has never been peer reviewed.
What the real study measured: 25 min 26 s, and not what you think
A published paper does exist, and it reports a neighboring figure. In 2005, Gloria Mark, Victor González, and Justin Harris presented at the CHI conference a direct observational study of 24 information workers, timed in the field. The result is unambiguous: when an interrupted task is resumed the same day, it is resumed after 25 minutes and 26 seconds on average.
But that duration does not measure the time needed to recover concentration. It measures the delay before returning to the task — the gap between the moment you drop it and the moment you reopen it. In between, as the authors state plainly, participants worked on an average of 2.26 other working spheres. Those 25 minutes are not 25 minutes of mental fog: they are 25 minutes during which you did two other things, often useful ones.
The distinction changes everything. "It takes 23 minutes to refocus" describes a cognitive handicap. "People return to an interrupted task after 25 minutes, having handled two others in between" describes how work is organized. The first framing invites guilt; the second invites you to look at your calendar.
The standard deviation nobody ever quotes
The same passage of the 2005 study carries a second number that gets stripped out of every retelling: the standard deviation of that resumption delay is 54 minutes and 48 seconds. It is more than twice the mean.
For a duration that cannot be negative, a standard deviation larger than the mean signals a heavily skewed distribution: a mass of quick returns, plus a long tail of tasks picked back up hours later that drags the average upward. The 25-minute "average" therefore describes almost nobody's actual experience. It is the arithmetic product of two very different populations of interruptions.
This is the same statistical trap as round health thresholds applied without qualification, such as the two-hours-of-screen-time-per-day rule: a convenient number travels all the better once it has shed the spread that made it honest.
In 2008, the same team found the opposite result
Three years later, Gloria Mark, Daniela Gudith, and Ulrich Klocke moved from the field to the lab, deliberately interrupting participants during an email-processing task. They expected to measure a cost in time. They measured the reverse.
Interrupted participants finished faster than the control group: 20.31 and 20.60 minutes depending on the condition, versus 22.77 minutes without interruption, with no difference in quality. The authors' interpretation is that people compensate: knowing they will be disturbed, they speed up and write less.
The cost is real, but it sits somewhere other than where everyone was looking. On the mental workload scales, the interrupted conditions scored significantly higher on stress, frustration, time pressure, and effort — after only twenty minutes of fragmented work. The price of interruption is not paid in lost minutes. It is paid in wear and tear.
What still holds once the myth is set aside
Dismantling a number does not dismantle the phenomenon. Three findings far more robust than the 23 minutes stand up perfectly well.
- One notification is enough, even ignored. Cary Stothart, Ainsley Mitchum, and Courtney Yehnert showed in 2015 that simply receiving a call or a text message significantly degrades performance on a sustained-attention task — without the participant touching the phone. The subjects did not even know the study was about notifications.
- Fragmentation has intensified. Gloria Mark has been measuring on-screen attention duration before switching since 2004. It was two and a half minutes in 2004, 75 seconds in 2012, and now hovers around 47 seconds.
- The current rhythm is documented. An Ipsos bva survey for Uside, fielded March 6–11, 2026 among 1,000 representative French employees, puts the rate at 4.7 interruptions per hour, or one every 13 minutes. Some 84 percent of respondents report concrete consequences on their work.
Notice how those numbers fit together: one interruption every 13 minutes, and a return to the original task after 25 minutes on average. The two do not contradict each other — they compound. You get re-interrupted before you have made it back.
The number that matters is your own
A 25-minute mean paired with a 55-minute standard deviation tells you nothing useful about your Tuesday. The right question is not "how long does it take to refocus," but "how long do I hold, and when." Four measurements are enough, and your devices already produce them.
Daily unlock count. Android and iOS count it without installing anything. It is the most honest proxy for self-interruption — the kind the 2006 interview already put at roughly 44 percent of the total. Nearly half of your interruptions come from you.
The length of your uninterrupted blocks. Not how many, how long. An afternoon of four 40-minute blocks and an afternoon of sixteen 10-minute blocks post identical total work time and have nothing in common.
The timing of your drop-offs. They almost always cluster into two or three stable slots. Identifying them beats imposing a generic rule on yourself: that is where a deliberate break costs the least, a point we develop in our article on micro-breaks and attention restoration.
Stress, not the stopwatch. That is the real lesson of the 2008 study. If your output holds but your resting heart rate climbs during fragmented weeks, interruption is costing you something — just not time. Cross-referencing those series is exactly what a dashboard that brings energy and productivity together makes possible, instead of tracking them in two apps that ignore each other.
None of these four measurements will produce anything as crisp as "23 minutes and 15 seconds." That is precisely what makes them useful: they describe an actual person. If the approach appeals to you but you are unsure where to start, our page on quantified self for wellbeing lays out the general framework.
FAQ
Is the 23-minute figure completely false?
It was not invented: Gloria Mark did state it, drawing on her own field data. What is false is its status and its meaning. It was never published in a peer-reviewed paper, and it refers to the delay before returning to an interrupted task, not the time required to regain concentration.
Why 23 min 15 s here and 25 min 26 s elsewhere?
They are two distinct figures. The 25 min 26 s appears in the study published in 2005, which also reports that 77.2 percent of tasks are resumed the same day. The 23 min 15 s comes from the 2006 interview, where the resumption rate given is 81.9 percent — a different dataset, one that was never published.
Do interruptions actually cost you time?
Not necessarily, and this is the most counterintuitive result in the literature. In 2008, the experiment by Mark, Gudith, and Klocke showed that interrupted participants finished faster, at equal quality, by picking up their pace. The cost moved elsewhere: more stress, frustration, time pressure, and perceived effort.
Does silencing your phone actually help?
The 2015 study on the attentional cost of notifications suggests the threshold is lower than most people assume: receiving a call or a message degraded performance even when participants never checked the device. Cutting the alerts themselves, and not only the time spent answering them, is therefore consistent with those results.
Where should I start measuring my own interruptions?
Start with your daily unlock count, which your phone already tracks. Record it for a week without changing anything, so you have a baseline. Then add the length of your uninterrupted work blocks: that duration, rather than total hours worked, is what separates a dense day from a fragmented one.