Self-Interruption Is Half the Story

Why We Act · 18 October 2026 · Reviewed by Faisal Zaidi

We carried the two-thirds figure for months, and it turns out to rest on one paywalled study of eighteen employees whose abstract never prints it. The number the record supports is half, and the question that follows is whether blocking the outside half simply moves it inside.

Answer to the question

What is self-interruption at work?

It is leaving a task before it is finished with nothing outside prompting you, and it accounts for about half of workplace interruptions rather than two-thirds. Gloria Mark's 2005 observation of twenty-four information workers found fifty-two per cent internal against forty-eight per cent external. A 2011 reanalysis of a thirty-six-person dataset counted 889 self-interruptions out of 2,030, which is about forty-four per cent.

Key takeaways

Gloria Mark's 2005 observation of twenty-four information workers split interruptions fifty-two per cent internal against forty-eight per cent external, not two-thirds to one-third.
A 2011 reanalysis of a thirty-six-person dataset counted 889 self-interruptions out of 2,030 interruptions, about forty-four per cent, and eighteen per cent of all task switches.
In the only randomised test, 247 people who disabled notifications for one working day did more deliberate self-checking, with no rise in involuntary internal interruptions that the study could detect.
Same-day resumption of interrupted work averaged twenty-five minutes and twenty-six seconds, not the twenty-three minutes usually quoted, with a spread more than twice the average.

Fifty-two per cent. That is the share of interruptions that information workers started themselves in Gloria Mark's observational study of twenty-four of them, against forty-eight per cent that arrived from someone or something else. So the useful rule is not a notification setting. It is this: before you leave a task, write the next physical action down, because self-interruption is often a memory event, and a note is cheaper than a return trip.

What should you do about it?

The advice in circulation is aimed at one half of the problem. Turn the alerts off, batch the email, close the door. Those are all moves against interruptions that arrive from outside, and roughly half of what breaks a working hour does not arrive from outside at all. It is started by the person doing the work. Four things follow from the evidence below, and they are the whole practical payload, so you can stop reading after the list.

Turn the notifications off anyway. In the one randomised test, deliberate self-checking did rise once the alerts stopped, at no measured cost, while the involuntary pull did not.
When something surfaces that you need to do later, write it down instead of going and doing it. Self-interruptions run towards work you are accountable for.
Budget for the return as well as the switch. Same-day returns to interrupted work averaged twenty-five minutes and twenty-six seconds, with a spread more than twice the average.
Look at the room before you look at your own discipline. Open-plan cubicles went with a much higher rate of self-interruption in one observational dataset.

The rest of this is why those four survive the evidence, and why a figure you have very likely read does not.

Where does the two-thirds figure come from?

Our own file carried a confident number for months: two-thirds of workplace interruptions come from the worker. It comes from The Distracted Mind, the 2016 book by Adam Gazzaley and Larry Rosen, at pages 113 and 114, where work episodes are described as lasting ten minutes or less and two-thirds of the interruptions to them as self-generated.

On the page, the figure is not attributed to the researcher most readers would assume. It is attributed to a shadowing study of eighteen employees at an Australian telecommunications company, led by Judy Wajcman. We went looking for that paper. It is paywalled everywhere we could reach, including the author's own institutional repository, which holds the record and not the text. Its abstract contains no such figure, and no citing source we found prints the percentage either.

So the honest position is that we could not read the number at source, and that nobody else in this literature reports a figure anywhere near it.

The field's own shorthand is half.

The field's own shorthand is half.

The fault sits in the chapter rather than in the authors, and it is worth being exact about where. Our audit found this the best-sourced book on our shelf, with about 350 numbered notes across its later chapters, about sixty-five per cent of them peer-reviewed, and critical papers printed next to supportive ones. The chapter carrying the two-thirds figure is the weakest in the volume at thirty-nine per cent peer-reviewed, and the same run of pages carries a vendor survey and a cost estimate our audit quarantined.

A single sociology paper on eighteen people in one firm is being asked to hold up a general fact about how all work is interrupted. That is a sourcing failure rather than a dishonesty, and it is the most common failure in this whole genre.

A single sociology paper on eighteen people in one firm is being asked to hold up a general fact about how all work is interrupted.

What do the observational counts actually show?

Three independent statements of the proportion sit in the published record. They cluster around half rather than near two thirds.

Fifty-two per cent internal against forty-eight per cent external, in Gloria Mark's 2005 observation of twenty-four information workers across more than seven hundred hours of watching. Managers skewed towards external interruptions; analysts and developers split about evenly.
About forty-four per cent, in Laura Dabbish's 2011 reanalysis of a thirty-six-person dataset: 889 self-interruptions out of 2,030 interruptions, which is eighteen per cent of all task switches once tasks that simply finished are counted as well.
"Half of the time they are interrupting themselves", the opening line of a 2009 conference paper by Jing Jin and Laura Dabbish, which is how the field states the figure to itself.

Not one of those is an estimate of a population.

Each is observation of information workers in a handful of technology, finance or research organisations, coded by somebody watching over their shoulder, and each is offered by its authors as a brief for designers rather than as a rate.

One further number needs correcting while we are here, because it travels further than the proportion does. The recovery figure in circulation is twenty-three minutes. The published figure, from that same 2005 study, is twenty-five minutes and twenty-six seconds for work resumed on the same day, and the spread around it is more than twice the average, so most returns are nothing like the average. About three-quarters of interrupted work came back the same day, and nine in ten of those returns were started by the worker rather than prompted by anybody.

Interruption also does not simply slow work down. In a laboratory study with forty-eight participants, most of them German university students, the interrupted versions of a simulated office email task were finished faster than the uninterrupted ones, with no difference in errors that the study could detect. The cost turned up as stress, frustration, time pressure and effort. Part of the speed was that people wrote shorter emails.

Does blocking reduce interruptions or move them?

Here is the question the whole argument turns on. If half of the interruptions are yours, does silencing the other half reduce the total, or does the pull simply reappear on the inside?

It has been tested properly once. Sandra Ohly and Luca Bastin randomly assigned 247 people, in work published in 2023, either to disable notifications as far as possible across every device for one working day or to leave their settings alone. Interruption frequency fell sharply in the group that switched them off, which is the manipulation doing its job. Deliberate self-checking rose. Internal interruptions, the involuntary pull to go and look, did not rise by anything the study could detect.

The distinction behind that result matters more than the headline. The authors separated checking you decide to do from the urge that arrives unbidden, and the data supported the split: internal interruptions went with lower performance and more irritation, while self-checking went with neither. What rose when the alerts stopped was the deliberate kind, at no measured cost.

They predicted two compensations and found neither. People high in fear of missing out, the worry that something is happening without you, were expected to interrupt themselves more once the alerts were gone. So were people high in telepressure, the felt obligation to answer messages quickly. Neither showed any association the study could detect.

One other result points the same way. In the 2011 reanalysis, external interruptions in one hour predicted a rise in self-interruption in the hour that followed, while self-interruptions did not predict more of themselves. The arrow in the measured data runs from outside to inside, which makes the outside the sensible place to push.

The arrow in the measured data runs from outside to inside.

The limits belong in the same breath as the finding. That randomised test ran for a single working day, measured internal interruption with one survey question, and recruited through personal networks in Germany, with more than half the sample students and three-quarters of them working mostly from home. Every measure was self-reported.

Nobody has run the test with behavioural measurement, and the gap is stated inside the literature rather than only by us. Gloria Mark's 2018 field study of thirty-two workers using blocking software says in its own introduction that research is lacking on how filtering potential distractors affects self-interruption, and that study counted no interruptions of either kind. An earlier experiment that cut email off entirely for thirteen workers counted window switches, which went down rather than sideways, and never measured self-interruption at all.

Why would anyone interrupt their own work?

The deficit reading is that self-interruption is a failure of attention. The measured pattern looks more like a function with a price attached.

In that 2011 dataset, people were more likely to interrupt themselves into a central working sphere, the term for a bundle of tasks worked towards a single goal, than into a peripheral one, and more likely to switch towards solitary work than towards talking to anyone. The authors read that as prospective memory, the ordinary business of remembering something you have to do. Self-interrupted work also took longer to come back to than externally interrupted work, although that particular difference was marginal.

What trips the switch is less settled. A laboratory study led by Rachel Adler gave 212 undergraduates six problem-solving tasks in separate tabs and let them move between them at will, and found negative feelings about progress triggering more switching than positive ones, with more switching going with lower accuracy.

The distribution is the part worth keeping. Of the 166 participants in the analysis, about a third switched for negative reasons, a little under a third for positive ones, and the remaining forty per cent reported no self-interruptions at all. Forty per cent of a sample doing none of it is not the shape of a universal deficit.

The seed book's own explanation sits one rung below that. Its chapter on why we interrupt ourselves imports a foraging model from behavioural ecology, and the authors flag the extension as their own proposal, repeatedly, so the diagrams in those pages are drawings of a hypothesis rather than plots of data.

Which leaves the two accounts on the shelf facing the same awkward number. The persuasive-design books argue that attention is taken from us, and have to answer a measurement saying half of it is handed over. The Distracted Mind reports the self-interruption figure and also writes that notifications exist because eyeballs bring money, and never reconciles the two. The design account deserves to be taken seriously, and we have taken it seriously in asking why you cannot stop scrolling, but the measured link from the external kind to the internal kind suggests the two are one story read at two points, and nobody on either shelf has written it that way.

It is also where this parts company with our work on friction. What cognitive load costs at checkout, the business of counting the steps in, then the steps out, and the sludge that keeps you subscribed are all about a step somebody added to a transaction, where the question is who the step serves, while here nobody has added a step and the person leaves the task on their own.

What can this evidence not settle?

The boundaries around all of the above are narrow, and they are worth stating before anybody builds a policy on it.

Every observational dataset here is office information work, in American, Mexican or German organisations, or else students in a laboratory. There is no shop floor, no clinical shift and no service counter in any of it. The two-thirds figure stays unverified at source, so treat it as a claim made in a book rather than as a finding. The open-plan result is an association in a dataset gathered in the early 2000s, and not evidence that the furniture causes anything.

The blocking question has one randomised answer, from one day, measured with one question. The commercial claim that blocking software improves work rests on two field studies with no control groups and self-assessed outcomes, whose own authors say as much in print. The lagged link from external interruption to self-interruption is a single coefficient in a four-page paper, on a dataset collected more than twenty years ago, that nobody has repeated.

So the four rules at the top are not readings off a settled science. They are the moves that survive the evidence as it currently stands, which is a weaker claim and the only one going.

If you keep one of them, keep the pen. Write the thing down the moment it surfaces, and let the task you were doing finish before you go and deal with it.

Common questions

What is self-interruption at work?

It is leaving a task before it is finished with nothing outside prompting you, which is roughly half of all workplace interruptions rather than two-thirds. Gloria Mark's 2005 observation of twenty-four information workers put the split at fifty-two per cent internal against forty-eight per cent external. A later reanalysis of a thirty-six-person dataset put it at about forty-four per cent.

Do most interruptions come from you?

About half do, on every observational count we could read at source. The three published statements of the proportion are fifty-two per cent, about forty-four per cent, and the field's own shorthand of half the time. The widely repeated two-thirds figure traces to one paywalled study of eighteen employees whose abstract contains no such number.

Does turning off notifications just move the interruption inward?

Partly, and not in the way the worry assumes. Sandra Ohly and Luca Bastin disabled notifications for 247 people for one working day. Interruptions fell, deliberate self-checking rose, and involuntary internal interruptions did not rise by anything the study could detect. Two predicted compensations, fear of missing out and telepressure, showed no association either. It was one day, measured by self-report.

How long does it take to get back to the interrupted task?

The published figure is twenty-five minutes and twenty-six seconds for work resumed on the same day, not the twenty-three minutes usually quoted. The spread around that average is more than twice the average itself, so most returns are much shorter or much longer. About three-quarters of interrupted work was resumed the same day at all.

Sources

González, V. M., & Mark, G. (2004). '"Constant, constant, multi-tasking craziness": managing multiple working spheres.' Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '04), 113-120. doi:10.1145/985692.985707 [Abstract read in full; author order confirmed at Crossref and OpenAlex]

Mark, G., González, V. M., & Harris, J. (2005). 'No task left behind? Examining the nature of fragmented work.' Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '05), 321-330. doi:10.1145/1054972.1055017 [Full text read at the author's institutional page; author order confirmed]

Jin, J., & Dabbish, L. A. (2009). 'Self-interruption on the computer: a typology of discretionary task interleaving.' Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '09), 1799-1808. doi:10.1145/1518701.1518979 [Abstract read in full; author order confirmed at OpenAlex. Full text not read]

Dabbish, L., Mark, G., & González, V. M. (2011). 'Why do I keep interrupting myself?: environment, habit and self-interruption.' Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '11), 3127-3130. doi:10.1145/1978942.1979405 [Full text read; author order and affiliations confirmed at Crossref]

Wajcman, J., & Rose, E. (2011). 'Constant Connectivity: Rethinking Interruptions at Work.' Organization Studies, 32(7), 941-961. doi:10.1177/0170840611410829 [Abstract read in full at Crossref, OpenAlex and the LSE Research Online record. Full text not available anywhere we could reach]

Adler, R. F., & Benbunan-Fich, R. (2013). 'Self-interruptions in discretionary multitasking.' Computers in Human Behavior, 29(4), 1441-1449. doi:10.1016/j.chb.2013.01.040 [Full text read at the open copy hosted by interruptions.net; author order confirmed at Crossref and OpenAlex]

Mark, G., Gudith, D., & Klocke, U. (2008). 'The cost of interrupted work: more speed and stress.' Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '08), 107-110. doi:10.1145/1357054.1357072 [Full text read; author order confirmed]

Mark, G., Voida, S., & Cardello, A. V. (2012). '"A pace not dictated by electrons": an empirical study of work without email.' Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '12), 555-564. doi:10.1145/2207676.2207754 [Full text read at the author's institutional page; DOI, page range and author order confirmed at Crossref]

Mark, G., Czerwinski, M., & Iqbal, S. T. (2018). 'Effects of individual differences in blocking workplace distractions.' Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems. doi:10.1145/3173574.3173666 [Abstract and method read in full from the author's institutional copy; author order confirmed from the paper]

Ohly, S., & Bastin, L. (2023). 'Effects of task interruptions caused by notifications from communication applications on strain and performance.' Journal of Occupational Health, 65(1), e12408. doi:10.1002/1348-9585.12408 [Full text read, open access at academic.oup.com; author order confirmed at the publisher record; PubMed 37280752]