Twenty-three minutes is real, and it is worse than the headline

The document has been open since ten and it is now half past two.

Somewhere in the last few years you met the number. Twenty-three minutes and fifteen seconds to get back to what you were doing, set in large type next to a photograph of somebody holding their head.

It is real. Somebody genuinely did stand behind office workers with a watch and time this. But it did not come from the paper you were sent, it does not measure the thing you were told it measures, and the sum people build on top of it does not fit inside a day.

Here is the number that is actually in the paper. Twenty-five minutes and twenty-six seconds, with a standard deviation of fifty-four minutes and forty-eight seconds.

Read the second half again, because it is the whole article. The spread is more than twice the average. That is not a tidy hill with a peak at twenty-five minutes. That is a heap of fast returns with a few disasters hanging off the end, dragging the mean upward. Your afternoon is not made of average interruptions. It is made of a great many cheap ones and one or two that quietly eat the rest of the day.

What the person with the stopwatch actually saw

The study is Gloria Mark, Victor González and Justin Harris, No Task Left Behind? Examining the Nature of Fragmented Work, presented at CHI in Portland in April 2005. Twenty-four information workers: seven managers, nine analysts, eight developers, drawn from two teams at two companies.

Each one was shadowed for three and a half days. The first half day existed so everybody could stop noticing the researcher standing there. The remaining three were timed to the second by a person with a watch and a notepad, who wrote down the moment the subject opened a document, dialled a number, started an email or turned around to talk to the next cubicle. That came to an average of 25 hours and 42 minutes of observation per person, and over 700 formal hours in total.

The unit they counted was not the task. It was the working sphere: a cluster of work that hangs together in your head. Getting the quarterly numbers out is one sphere. The email about the quarterly numbers sits inside it. The unrelated email that arrives while you are in it does not.

Three findings are worth carrying around for the rest of your career:

• People were holding an average of 11.7 working spheres at once.
• They stayed in one for an average of 11 minutes and 4 seconds before switching or being interrupted.
• 57.1% of those stretches ended in an interruption rather than in being finished.

Eleven minutes is the honest length of the thing you are defending at your desk. Not the ninety-minute deep work block, not the tomato timer. Eleven minutes, and then something happens. Which is worth knowing before you pick how long your focus blocks should be, because most people pick a length the day has never once given them.

The twenty-three is in an interview, not in the paper

People have gone looking for "23 minutes 15 seconds" in the academic literature, come back with nothing, and concluded the statistic was made up. It was not made up. It is simply in a different kind of document.

The source is a Gallup Business Journal interview with Gloria Mark published in June 2006, where she gives 23 minutes 15 seconds as the time to resume interrupted work, and says in the same breath that people did roughly two other things first.

That interview describes thirty-six workers. The paper describes twenty-four. So the figure everybody quotes is drawn from a wider slice of her fieldwork than the single paper everybody attaches it to, which is precisely why searching that paper for it comes up empty and why a small internet genre now insists the whole thing is a myth.

Twenty-three or twenty-five, the argument does not turn on which you use. It turns on one word that gets dropped every single time.

The word that gets dropped is "elapsed"

Twenty-three minutes is elapsed clock time between leaving a piece of work and touching it again. It is not twenty-three minutes of fog. Nothing in that measurement says anything whatsoever about the state of your brain, because a person with a stopwatch cannot see the inside of your head.

During those minutes you were working. The paper puts a number on that too: before returning, subjects passed through an average of 2.26 other working spheres. Two more real pieces of real work, done properly, in between.

So the honest sentence is this. When you leave a thing, you typically do about two other things and come back around twenty-five minutes later, having been busy throughout. Which is a fair description of most people's Tuesday, and is the reason the day can feel completely full and completely wasted at the same time.

You did not lose twenty-three minutes. You spent them, on two other things, in an order you did not choose.

The sum everybody builds on it runs past midnight

Take the study's own figures and do the arithmetic people do out loud in meetings, where every interruption costs you twenty-three minutes of your life.

Ten hours of recovery inside a day that was eight and a half hours long. On that reading, those twenty-four people spent every working minute recovering from work they were also doing, and mostly finishing, while a researcher watched.

The sum is not slightly out. It is out by more than a working day. Any time you see "four interruptions this morning, so that is an hour and a half gone", that is this sum, and it is this broken.

Three things in the data that are genuinely worse

You are the one doing it, and it costs you more when you do

Across the 1,282 interruptions in the paper's table, 52% were internal. Self-started. Nobody walked over, nothing rang. The person simply left.

And then the finding that never makes the poster. Work interrupted by somebody else came back in 22 minutes and 37 seconds. Work people interrupted themselves out of came back in 29 minutes and 1 second. Six and a half minutes worse when it was your own idea.

Being straight about it: the paper reports that gap at p = .055, which sits just the wrong side of the line researchers usually draw. It would be a poor article about ignored caveats that then quietly ignored one. Treat it as a strong hint rather than a settled result. Every other number here points the same way.

Roughly a quarter of it never comes back at all

77.2% of interrupted work was resumed the same day. Which means just under a quarter of it was not.

And the twenty-five-minute average is calculated only over the work that came back, because you cannot time a return that never happens. It is a survivor's average. Every piece of work that was abandoned that day sits outside the number entirely, and that is the one place where the headline is being too kind to you.

The real cost sits on both sides of the gap, not inside it

Sophie Leroy's 2009 work on attention residue found that part of your attention stays with the thing you left, especially when you left it unfinished, and that your performance on the next thing measurably suffers for it.

Mark's own lab study three years later found something stranger. People finished interrupted work faster, with no loss of quality, while reporting more stress, more frustration, more time pressure and more effort. Interruption did not make them slower. It made them finish in a worse state. That is a genuine cost, and no stopwatch will ever see it.

So count departures, not minutes

You cannot count your recovery minutes. You would need somebody standing behind you with a watch, and even then the number would describe your day rather than change it.

You can count departures. Your phone has been counting them since 2018 and has never once mentioned it.

Settings › Screen Time › See All App & Website Activity, then scroll past the app list to Pickups. Two numbers live there: the total, and First Used After Pickup.

Ignore the total. It counts screen wakes, including the ones a notification caused and the ones where you only wanted the time, so it flatters nobody and measures nothing you decided. The number that matters is First Used After Pickup: which app your thumb went to first, and how many times it did that yesterday.

That is your count of self-started departures. Look at the top row. That is not a statistic about office workers in 2005. That is you, yesterday, and it is almost certainly one app doing most of the work. If you cannot say what you wanted from it when you reached, you did not really open it for anything.

Now the arithmetic that actually works. The study's day held about forty-six chances to leave, and about half the leaving was the person's own idea. You cannot uninstall your colleagues. Slack is your actual job. The fire alarm is allowed to interrupt you. What remains is your half, and most of your half is one app, reached for with one thumb, in well under a second.

There is nothing between the reach and the app

That is the whole problem with the reach, and it is what makes it different from every other interruption in the study. The others all had visible machinery: somebody walked over, a phone rang, a meeting started, a door opened. The reach has no machinery at all. Intention to open app is a single motion, and it completes before the part of you that would have objected has finished forming the objection.

A timer does not help with that, because a timer is information and you already have the information. What helps is something standing in the gap. Something that has to be got past, that is not you, and that was not there when you decided you wanted this.

Which is where we admit what we make. Scrappy is a small dog who stands in front of the apps you name. He knows three rules and refuses to learn a fourth: quiet hours, a daily limit, and an open limit, which is the one this whole article has been about. Six looks a day is a rule about departures rather than minutes, and departures are the thing the stopwatch actually found.

First knock, he is polite about it. Fourth knock, he is standing on the doorknob with an opinion. Getting past him costs a treat, and treats come out of real life: about two thousand steps of walking earns one, and one buys five honest minutes. You cannot buy them with money. We will never sell them, at any price, in any version.

And when the work needs a proper stretch, you take him out. A walk is a named session, twenty-five minutes to eight hours, with a Live Activity on the lock screen, so for once the phone is announcing what it is doing instead of the other way around.