Attention and habit
Interruptions, focus and the numbers
The two most-quoted statistics about interruption have no peer-reviewed paper behind them, and the published experiment cited next to one of them found the opposite.
The two numbers that dominate this subject have the least behind them. "It takes 23 minutes and 15 seconds to get back to the task" comes from a magazine interview, not a paper, and has never been published with a sample size or a confidence interval. "Multitasking costs up to 40% of productive time" is a verbal estimate by a psychologist; the switch costs in the experiments underneath it were sometimes a few tenths of a second. What has been measured is less quotable: interruption makes people work faster and feel worse, notifications alone raise error rates in the lab, and the association between media multitasking and cognitive control is close to nothing once self-ratings are set aside. For a workplace-oriented comparison point on how digital activity is translated into metrics, see task switching cost.
The 23-minute figure is an interview, not a finding
The trail ends in a trade-press interview with Gloria Mark of UC Irvine.
The "23 minutes and 15 seconds to get back to the task" figure comes from a 2008 Fast Company interview with Gloria Mark, not from a peer-reviewed paper reporting that result.
Direction: Cuts both ways. Strength of evidence: Strong.
Caveat The underlying quantity is real observational data from ethnographic workplace studies, but it has never been published with a sample size, distribution or confidence interval in the form quoted.
The figure is not fabricated — it comes from a serious researcher describing her own naturalistic data. But it cannot bear the weight put on it, because no distribution has been published: no spread, no task type, no indication of whether the median resembles the mean. It is often attributed to a 2005 CHI paper by Mark and colleagues; that attribution has not been verified against the paper itself and should not be repeated without checking the original.
The published experiment found the opposite of the implication
The peer-reviewed study cited alongside the 23-minute claim measured interrupted and uninterrupted work directly.
Interrupted tasks were completed faster (20.3–20.6 minutes) than uninterrupted ones (22.8 minutes), but with significantly higher stress, frustration, time pressure and effort.
Direction: Decrease. Strength of evidence: Strong.
Caveat A lab email-answering task with 48 mostly student participants, not sustained knowledge work in real conditions.
People compensate for interruption by working faster and pay for it in stress rather than elapsed time — a more interesting finding than the one it is used to support.
The 40% figure is a quotation
The other pillar comes from an American Psychological Association explainer page.
The claim that multitasking costs up to 40% of productive time is a verbal estimate attributed to psychologist David Meyer, while the switching costs actually measured in the underlying experiments were sometimes just a few tenths of a second per switch.
Direction: Decrease. Strength of evidence: Weak.
Caveat Extrapolating from sub-second lab switch costs to 40% of a working day is an unpublished inference, and the APA page presents the figure as a quotation from Meyer rather than as a study result.
Sub-second costs are genuine and they accumulate. Going from a few tenths of a second on arithmetic and shape sorting to two days a week of working life requires assumptions nobody has published — a leap made easier by the difficulty of measuring productivity at all.
Notifications cost something even when ignored
One lab demonstration kept the phone out of the participant's hands entirely.
Merely receiving a notification, without looking at or responding to the phone, more than tripled the probability of errors on a sustained-attention task, with calls more disruptive than texts.
Direction: Increase. Strength of evidence: Mixed.
Caveat A single lab study on a monotonous task in which participants did not expect to be contacted, which maximises surprise and probably overstates the cost of routine notifications.
Removing the source of interruption changes behaviour measurably. When 13 information workers at one organisation had their email cut off for five days, window and task switching fell from 37.1 to 18.2 an hour and average time in a single window more than doubled, from 75.5 to 131.9 seconds (Mark, Voida & Cardello, CHI 2012); the heart-rate-variability signal of reduced stress did not reach conventional significance at p = .065. It is one small within-subjects study at a single organisation, but it is a rare direct measurement.
The multitasking-harms-cognition case has weakened
A decade of studies after the influential 2009 finding produced a much smaller effect than the original.
The association between media multitasking and cognitive control is r = .138 overall, falling to r = .091 for performance-based measures against r = .200 for self-report measures.
Direction: Increase. Strength of evidence: Strong.
Caveat The authors conclude that the field is no closer to understanding cognitive control in media multitaskers; the effect survives mainly where both variables are self-reported, the classic signature of common-method bias.
Self-assessment is not a usable substitute either: 70% of 277 participants rated their own multitasking ability as above average, yet perceived ability was essentially uncorrelated with measured ability at r = .08 and frequent multitaskers scored lower on executive control (Sanbonmatsu, Strayer, Medeiros-Ward & Watson, PLOS ONE, 2013).
The related claim that a phone on the desk drains cognitive capacity has fared badly under direct replication.
A preregistered replication of the "brain drain" effect found no difference in operation-span or go/no-go performance whether the phone was on the desk, in a pocket or bag, or outside the room, on or off.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat A meta-analysis of 22 studies (43 effects, about 3,558 participants) does find a small pooled effect of g = −0.14, driven by memory tasks and by Asian samples, with essentially nothing in North American samples — so the field is genuinely split.
Attention spans at the screen have shortened
The one trend measurement with reasonable support comes from computer logging of information workers, though it is reported through interviews and a trade book rather than a single paper.
Average attention on a single screen fell from about 2.5 minutes in 2004 to 75 seconds in 2012 and to about 47 seconds recently, with independent replications finding 50 seconds, 44 seconds and a median around 40.
Direction: Decrease. Strength of evidence: Mixed.
Caveat Reported via interview and trade book rather than a peer-reviewed paper with pooled sample sizes, and the 2004-to-now comparison runs across different studies with different logging tools, not one cohort followed over time.
That figure sits comfortably alongside the logged evidence on how fragmented ordinary phone use is, where logged sessions mostly run under seven minutes (Toth, Parry & Emmer, Journal of Quantitative Description: Digital Media, 2025).
The short version
- The 23-minute refocus figure originates in a 2008 Fast Company interview and has never been published with a sample size or confidence interval.
- The peer-reviewed experiment usually cited with it found interrupted tasks were finished faster, at 20.3 to 20.6 minutes against 22.8, but with more stress and frustration (Mark, Gudith & Klocke, CHI 2008).
- The 40% productivity-loss claim is a quotation attributed to David Meyer, extrapolated from switch costs measured in tenths of a second (APA, citing Rubinstein, Meyer & Evans, 2001).
- Receiving a notification without touching the phone more than tripled error rates on a lab attention task (Stothart et al., 2015), though the design maximised surprise.
- The evidence on media multitasking and cognitive control is weak and probably inflated by method: the pooled association is r = .138, and only r = .091 when performance is measured rather than self-rated.