The Borrowed Premise

Answer to the question

Is there any evidence that deep work improves output?

Newport does not argue the premise of deep work. At page seven he names four other people's books and writes that he will not spend more time establishing the point. The argument built on that loan is useful, but we searched twice and found no field study showing that protected blocks of time produce better or more work.

Key takeaways

• On page seven of Deep Work the founding premise is not argued but borrowed, from four trade books by three other authors, with Newport writing that he will not spend more time establishing the point.
• The twenty-three minute figure is Gloria Mark's own, given in a Gallup interview of 8 June 2006 for the 81.9 per cent of work resumed the same day, and glossed by her as not so long.
• The published figure for that measure, from her 2005 CHI paper on twenty-four workers at one firm, is twenty-five minutes and twenty-six seconds, with a standard deviation of fifty-four minutes and forty-eight seconds.
• We searched twice for a field study showing that blocks of uninterrupted time produce better or more work and found none, and the argument of the book does not depend on any of the citations that failed.

The question people ask about deep work is whether they are getting enough of it. That assumes the answer. The better question is what an uninterrupted hour measurably produces.

Here is where we land. Cal Newport's Deep Work does not argue its own founding premise: on page seven it points at other people's books and declines to. The argument he builds on that loan is genuinely useful. The famous number attached to the subject appears in no paper we could locate, and we searched twice for field evidence of the productivity claim without finding it.

Newport, on page seven, declining

On page seven of the Introduction, read here from photographs of the UK paperback, Newport names Nicholas Carr's 2008 Atlantic article and four trade books: Carr's The Shallows, William Powers's Hamlet's BlackBerry, John Freeman's The Tyranny of E-mail and Alex Soojung-Kim Pang's The Distraction Addiction. That is four books by three authors, Carr counted twice. Then comes the sentence the whole structure turns on:

> "Given this existing body of evidence, I will not spend more time in this book trying to establish this point."

The move is to ask the reader to grant that network tools damage the capacity for concentration, so the book can get on with the prescription. Grading a book on a premise it declines to defend is a choice, and calling that decline the most consequential methodological decision in the introduction is our judgement rather than a finding. The premise is never defended later either. Rule 3 opens at page 182 by declining it again, on the grounds that fragmentation "no longer generates much debate; we all feel it".

Four pages earlier, the load-bearing science claim arrives bare. "We now know from decades of research in both psychology and neuroscience that the state of mental strain that accompanies deep work is also necessary to improve your abilities," he writes at page three, naming no researcher and no study. The Notes run from page 265 to page 287, and counting the Introduction's entries there by hand gives fifteen, of which not one is a peer-reviewed source. The stronger claim on page seven, that enough frenetic shallowness permanently reduces the capacity for deep work, carries no source either.

The twenty-three minute figure, the most quoted number in the attention business, appears nowhere in his book. His single use of Gloria Mark sits at page fifty-two, credited to Mark and her co-authors with the study unnamed, and the endnote resolves to the correct paper.

Carr, whose best evidence later failed

Of those four books, this publication has read exactly one. We grade Carr's The Shallows contested, for reasons already published. The study he added in his 2020 afterword as fresh vindication, on the cognitive cost of a phone sitting on the desk, largely fails meta-analytically: a synthesis of 166 effect sizes from 33 studies, with more than four thousand participants between them, finds an effect indistinguishable from zero.

Our verdict on his book is that the narrow claim about screens is real and bounded, while the general degradation of attention and memory is not currently evidenced, and its best-known support has failed. The other three have never been read here, so we have no verdict on them.

A borrowed premise inherits whatever happens to the lender, and in 2016 nobody could have known what would happen to this one. Our hub records what survives in Carr, the historical argument that deep reading is a built capacity, and records our objection to our own position: this literature measures what is easy to measure, and a slow change in a habit of mind is close to unmeasurable.

A borrowed premise inherits whatever happens to the lender, and that is the cost of the loan.

Mark, in her own words

The figure everyone quotes comes from an interview. On 8 June 2006, Jennifer Robison asked Gloria Mark how long it takes to get back to work after an interruption. Mark's answer, read here in full, opens "There's good news and bad news", explains that she looked only at work interrupted and resumed on the same day, puts that same-day share at 81.9 per cent, and gives the number:

> "it was resumed, on average, in 23 minutes and 15 seconds, which I guess is not so long."

The person who measured the quantity offered it as the good news. The bad news, in her framing, is the detour: about two intervening tasks arrive before you return, so reorienting takes more effort. Gallup's own standfirst reads "Interruptions can actually be quite beneficial", and Mark describes an interruption on the topic you are already working on as an interaction, which the workers she shadowed found helpful. The piece that launched the number argues close to the opposite of what the number is used to argue.

The most quoted number in the attention business was offered by the person who measured it as the good news.

Pressed on whether there must be a cost to productivity, she answered: "There must be, but we don't know what [it is]." That was twenty years ago. The interview also describes a larger dataset than any paper we could reach, thirty-six managers, analysts, developers, engineers and project leaders shadowed for three days, and we searched again without locating a paper that carries it.

The paper, as actually published

The published source is Mark, with Victor Gonzalez and Justin Harris, at the CHI conference in 2005, read here in full. "When people did resume work on the same day, it took an average length of time of 25 min. 26 sec," the paper reports, with a standard deviation of fifty-four minutes and forty-eight seconds. The spread is more than twice the mean, so most returns look nothing like the average, a point [Self-Interruption Is Half the Story](https://thewhyweproject.com/articles/self-interruption-is-half-the-story) has already made.

What was measured is narrower than the circulated version suggests: elapsed clock time between an interrupted working sphere and the worker's return to it, same day only, with the final hour of the day excluded. The method was stopwatch ethnography of twenty-four information workers at one outsourcing firm serving financial bond managers, over 700 formal hours, no video permitted. The authors restrict their own generalisation to high-pressure firms with many integral tools.

The circulated version therefore departs from the published one on four separate counts.

- The number. Twenty-five minutes and twenty-six seconds in the paper, twenty-three minutes and fifteen seconds in the interview, with no paper we could locate carrying the second one.
- The condition. Both figures cover only work resumed the same day, 77.2 per cent in the paper and 81.9 per cent in the interview, so each understates the average cost across all interrupted work.
- The dispersion. A standard deviation of fifty-four minutes and forty-eight seconds drops out, and a very wide distribution becomes a flat point estimate.
- The setting. Twenty-four people at a single firm, with the working sphere as the unit of analysis, becomes a fact about how long it takes anybody to refocus.

The fault lies in the circulation rather than with the researcher, who conditioned the figure, published the larger one and called it short in the same sentence. She called it short for the reason given above: the cost she locates is the reorienting across the two intervening working spheres, which puts her nearer Newport's position than the usual framing of this disagreement allows.

A further published detail is often read as cutting the other way, and it will not bear it. Externally interrupted work came back in about twenty-two and a half minutes, self-interrupted work in about twenty-nine, but that difference did not reach conventional significance, falling just the wrong side of the one-in-twenty threshold on a comparison of nearly a thousand resumptions. It cannot carry the inference that the slower half of the famous number is the half people inflict on themselves. The same paper found that workers sitting together were interrupted more and worked longer stretches, which is the territory of [The Office That Stopped People Talking](https://thewhyweproject.com/articles/the-office-that-stopped-people-talking).

Mark and Gudith, on interrupted workers

Mark's own experiment came three years later, with Daniela Gudith and Ulrich Klocke, and it is the nearest thing to a direct test of the deep work case. Forty-eight subjects, mostly German university students, played a human resources manager back from holiday and answered twelve emails per condition, with interruptions arriving every two minutes. The interrupted conditions finished faster than the undisturbed one, with no measurable difference in errors or in rated politeness. Our [attention residue](https://thewhyweproject.com/articles/why-attention-residue-lingers-at-work) piece takes the rest. Their abstract puts it plainly:

> "people completed interrupted tasks in less time with no difference in quality. Our data suggests that people compensate for interruptions by working faster, but this comes at a price: experiencing more stress, higher frustration, time pressure and effort."

Two qualifications ride along, both supplied by the authors. Part of that speed is shorter writing, since emails were significantly longer in the undisturbed condition, which they tested themselves. They also raise the possibility that interruptions genuinely lengthen a task, with the extra time falling in the moments just after the interruption, masked by a faster and more stressful style. "More sophisticated measurements of working speed directly after an interruption must be done to test this," they wrote, in 2008.

Where the cost did show up was in the person. Stress, frustration, time pressure, effort and mental workload all rose on scales running to twenty, after as little as twenty minutes of interrupted performance, and the authors cannot tell whether people would cope over a longer stretch. We have used that study before and cite it as an established house finding.

Perlow and Das Swain, on output

The deep work prescription rests on Leslie Perlow's 1999 field study of one software team, at a company she calls Ditto, where she ran quiet time in three phases. Read in full, its outcome measure is stated in her own words: questionnaire data from the seventeen members of the group, including self-ratings of their productivity against the period before the study. The daily tracking logs, the one objective instrument, were abandoned when the engineers stopped filling them in accurately, and she declines to report them.

The shares rating themselves above average ran 59, 41 and 65 per cent across the three phases, then 47 per cent a month after the last ended. The hard outcome is one executive's testimonial: a division vice president later credited the study with the product's on-time launch. Nine months on, Perlow found the lasting effects minimal, and a year later one engineer told her that quiet time alone does not work and nobody kept to it.

The best randomised test of what the trade actually sells is much newer. Das Swain and colleagues, at CHI in 2023, gave calendar-protected focus blocks to 89 workers who had never had them, with one baseline week, a three-week intervention and 48 people randomised into treatment, in engineering roles at one large technology company. Treatment workers self-reported higher performance, at a level the paper itself describes as significant only within a ninety per cent confidence interval. The measure closest to output, daily engagement on coding and development, moved slightly downward, though not significantly. That study is the top rung of the ladder, and it has the same hole in it as every rung below.

Study	Design	What was measured	Direction
Das Swain and colleagues 2023	Randomised, 89 workers, three weeks	Self-rated performance	Gain at the ninety per cent level; coding activity slightly down
Chen and Chan 2026	Randomised, 217 participants, four weeks	Work time, hours at the work	Complete blocking backfires; partial blocking helps
Marotta and Acquisti 2018, a working paper	Randomised, piece-rate crowdworkers	Tasks completed per hour	More tasks when imposed, nothing when chosen
Pendem and colleagues 2022	Instrumental variable, 211 harvesters	Truckloads of fruit	Some interruptions raise output
Vasilescu and colleagues 2016	Correlational, 1,193 developers	Lines of code per week	An inverted U; within-day switching positive
Perlow 1999	Collaborative intervention, seventeen engineers	Self-rated productivity	Short-term gain, gone inside a year

We looked for better twice, in extended searches, for meeting-free day evaluations with a control group and for focus-time interventions measured against output. Nothing came back from either pass that measures output. Those were targeted searches of the reachable published and preprint record, so the honest form of the claim is that none has been found.

Two entries deserve saying aloud. The fruit-harvesting result is the best-powered objective-output study of interruption in existence, and its beneficial interruptions supply physical respite, a mechanism that does not transfer to a desk. The developer data points several ways at once, with focus helping at the week level and within-day switching associated with slightly more code. Whether an absorbed state can be located at all is settled in [What the Flow Studies Measured](https://thewhyweproject.com/articles/what-the-flow-studies-measured).

Our verdict, on a useful argument

Our verification pass is unambiguous about the half that holds. The central practical claim, that the ability to concentrate without interruption is both valuable and increasingly rare, does not depend on any of the citations that failed. A book can be badly sourced and still be right, and on its main thesis it is. The apparatus contains no famous-name misattribution anywhere in twenty-three pages of notes, which in this genre is unusual.

There is more to credit. Newport attributes "don't break the chain" to Brad Isaac at Lifehacker instead of to Jerry Seinfeld, the one place in our whole vault where a book got the popular attribution right. At pages 130 to 133 he takes a historical case that looks like a counterexample, goes to the actual floor plans of Building 20 and Bell Labs, finds private offices on shared corridors, and builds his prescription out of what he found. And in 2021, in A World Without Email, he argued that the problem is organisational, which is the structural reading Deep Work had little room for in 2016.

What he concedes at page seventy is the hinge. There are no hard metrics relating either depth or shallowness to the bottom line, he writes, naming this the metric black hole. The same absence applies to his own claim that depth would outperform, and that symmetry goes unmentioned. What follows is an inference chain: depth is hard, shallowness is visible, nothing can be measured, therefore depth is rare, therefore rare is valuable, therefore go deep.

The second-most-quoted number in the deep work trade repeats the pattern. The claim that multitasking costs 40 per cent of productive time is not in Rubinstein, Meyer and Evans's 2001 task-switching paper, whose measured costs run from 29 to 975 milliseconds and whose retrieved text, of which we read about 80 per cent, mentions productive time nowhere.

Nor is it in the 2001 press release. It exists in one sentence on an American Psychological Association topic page created in 2006, recording that Meyer has said brief mental blocks can cost as much as 40 per cent of someone's productive time. An upper bound on a conditional, offered informally, hardened into a point estimate of a mean across twenty years of citation, and the error belongs to that chain.

Each number was accurate when it was said, Mark's conditioned on same-day resumption and Meyer's hedged with "as much as", and both reached the trade stripped of the qualifier that made it true. So the title earns itself twice over: a careful writer borrowed a premise he could not audit from books whose best-known evidence has since collapsed, and the trade borrowed a careful researcher's conditioned figure and lost the condition somewhere along the way, which is how a book scrupulous about where it got everything came to rest on a mechanism nobody ever checked.

Common questions

Is there any evidence that deep work improves output?

Not in the field evidence we could reach. We searched twice for a study showing that blocks of uninterrupted time produce better or more work, and found none. The best randomised tests measure self-rated performance or hours at the work, the only randomised study with a real output measure is an unpublished working paper, and the one study with an objective output measure, on fruit harvesters, is an instrumental-variable design rather than a clean causal test.

Where does the twenty-three minute figure come from?

From a Gallup interview with Gloria Mark published on 8 June 2006, where she gave it for the 81.9 per cent of interrupted work resumed the same day and added that it was "not so long". Her published paper, at CHI in 2005, reports twenty-five minutes and twenty-six seconds for the same measure, with a standard deviation of fifty-four minutes and forty-eight seconds.

Does Cal Newport argue the premise of Deep Work?

He declines to, explicitly. At page seven he names a magazine article and four trade books by three other authors, writes that he will not spend more time in the book trying to establish the point, and moves on to the prescription. Rule 3 declines it a second time at page 182. The deep work argument built above that premise does not depend on it.

Sources

Newport, C. (2016). Deep Work: Rules for Focused Success in a Distracted World. Piatkus. ISBN 978-0-349-41190-3

Mark, G., Gonzalez, V. M., & Harris, J. (2005). 'No Task Left Behind? Examining the Nature of Fragmented Work.' Proceedings of CHI 2005, 321-330

Robison, J. (2006). 'Too Many Interruptions at Work?' A Q and A with Gloria Mark. Gallup Business Journal, 8 June 2006

Mark, G., Gudith, D., & Klocke, U. (2008). 'The Cost of Interrupted Work: More Speed and Stress.' Proceedings of CHI 2008, 107-110. doi:10.1145/1357054.1357072

Perlow, L. A. (1999). 'The Time Famine: Toward a Sociology of Work Time.' Administrative Science Quarterly, 44(1), 57-81. doi:10.2307/2667031

Das Swain, V., Hernandez, J., Houck, B., Saha, K., Suh, J., Chaudhry, A., Cho, T., Guo, W., Iqbal, S. T., & Czerwinski, M. (2023). 'Focused Time Saves Nine: Evaluating Computer-Assisted Protected Time for Hybrid Information Work.' Proceedings of CHI 2023. doi:10.1145/3544548.3581326

Chen, Z., & Chan, J. (2026). 'Less Is Not Always More: The Impact of Social Media Block Intensity and Immediacy on Work Time.' Information Systems Research. doi:10.1287/isre.2023.0493

Marotta, V., & Acquisti, A. (2018). 'Interrupting Interruptions: A Digital Experiment on Social Media and Performance.' SSRN working paper 3283951. doi:10.2139/ssrn.3283951

Pendem, P. K., Green, P. I., & Staats, B. R. (2022). 'The Microstructure of Work: Understanding Productivity Benefits and Costs of Interruptions.' Manufacturing and Service Operations Management, 24(4), 2202-2220. doi:10.1287/msom.2021.1053

Vasilescu, B., Blincoe, K., Xuan, Q., Casalnuovo, C., Damian, D., Devanbu, P., & Filkov, V. (2016). 'The Sky Is Not the Limit: Multitasking Across GitHub Projects.' Proceedings of ICSE '16. doi:10.1145/2884781.2884875

Puranik, H., Koopman, J., & Vough, H. C. (2020). 'Pardon the Interruption: An Integrative Review and Future Research Agenda for Research on Work Interruptions.' Journal of Management, 46(6), 806-842. doi:10.1177/0149206319887428

Rubinstein, J. S., Meyer, D. E., & Evans, J. E. (2001). 'Executive control of cognitive processes in task switching.' Journal of Experimental Psychology: Human Perception and Performance, 27(4), 763-797. doi:10.1037/0096-1523.27.4.763

American Psychological Association (2001). 'Is Multitasking More Efficient? Shifting Mental Gears Costs Time, Especially When Shifting to Less Familiar Tasks.' Press release, 5 August 2001

American Psychological Association. 'Multitasking: Switching costs.' Topic page, apa.org/topics/research/multitasking, created 2006

Leroy, S. (2009). 'Why is it so hard to do my work? The challenge of attention residue when switching between work tasks.' Organizational Behavior and Human Decision Processes, 109(2), 168-181. doi:10.1016/j.obhdp.2009.04.002

Ward, A. F., Duke, K., Gneezy, A., & Bos, M. W. (2017). 'Brain Drain: The Mere Presence of One's Own Smartphone Reduces Available Cognitive Capacity.' Journal of the Association for Consumer Research, 2(2), 140-154. doi:10.1086/691462

Related concepts

• Citation Asymmetry
• The Stipulated Premise
• The Metric Black Hole

People

• Nicholas Carr
• David Meyer