A message flashes in the corner of a screen. Answering it takes perhaps 20 seconds. Yet a widely repeated finding from workplace research suggests that people may not return to the interrupted task for roughly 23 minutes.
That number sounds like a stopwatch on the mind, but it needs careful translation. The researchers measured when people resumed a piece of work, not when they had fully recovered an earlier depth of concentration. The distinction matters because “back on the task” and “thinking as deeply as before” are not the same outcome.
The result also came from a small observational study of 24 information workers at one company, conducted before Slack and smartphone notifications became ordinary. This is one study, not settled consensus. It remains useful because it shows what can fill the interval after an interruption: usually not 23 minutes of staring blankly, but a chain of other tasks.
The 23-minute figure came from watching real work
In a 2005 paper presented at the ACM CHI conference, Gloria Mark, Victor Gonzalez and Justin Harris described shadowing managers, financial analysts and software developers at an outsourcing company. Researchers recorded activities to the second, noting when someone opened a document, wrote an email, made a call or spoke with a colleague.
The team grouped those events into “working spheres,” meaning larger projects or themes that could involve several applications and conversations. Across the sample, 57 percent of working-sphere segments were interrupted. About 77 percent of interrupted work was resumed on the same day.
When same-day resumption happened, the average delay was 25 minutes and 26 seconds. Externally interrupted work returned sooner, after an average of 22 minutes and 37 seconds, while self-interrupted work took 29 minutes and one second. The familiar 23-minute-and-15-second statistic is a rounded public summary from Mark’s broader interruption research, sitting close to the paper’s result for external interruptions.
There was enormous variation. The standard deviation for the overall resumption delay was nearly 55 minutes, a sign that the average did not describe every interruption neatly.
The study measured resumption, not restored focus
The paper did not use a test of concentration before and after each interruption. It did not measure whether someone had entered a “flow state,” nor did it determine the moment at which reasoning quality returned to its earlier level. There were no brain scans, attention scores or physiological markers of deep focus.
Its 25-minute measure was simpler: the elapsed time before a person resumed the interrupted working sphere. That could mean reopening a document, continuing an analysis or returning to a project after several unrelated activities.
This makes the finding less dramatic than the viral version, but in another sense more informative. On average, participants entered 2.26 other working spheres before returning. The cost was not a fixed mental recovery period. It was the way one interruption created room for two more pieces of work, each carrying its own context and demands.
A text message and a Slack ping are reasonable modern examples of external interruptions, but neither was directly tested in that 2005 study. Its recorded interruptions included phone calls, colleagues entering a workspace and email alerts. Different technologies can create different patterns, even when the underlying act of leaving one task for another looks familiar.
People often switch again before 23 minutes have passed
The same research program found that work was already fragmented on much shorter timescales. The 2005 paper cited an earlier observation in which information workers switched work events about every three minutes. In the later sample, people generally stayed within a broader working sphere for around 10 to 12 minutes before switching or being interrupted.
Neither figure means that every person receives a notification at that exact interval. Some switches are voluntary. Others occur when a task ends naturally, a colleague needs an answer or a more urgent responsibility appears. The authors found that analysts and developers interrupted themselves about as often as outside events interrupted them, while managers experienced more external interruptions.
More recent work by Mark uses another unit entirely: time spent on one screen before shifting. In a 2023 University of California, Irvine interview, she said that this measure had fallen from about two and a half minutes in 2004 to an average of 47 seconds in more recent datasets. Screen switching, task switching and interruption are related ideas, but they should not be treated as interchangeable measurements.
The study often cited for 23 minutes tested something else
Online accounts frequently link the 23-minute claim to a different 2008 CHI paper by Mark, Daniela Gudith and Ulrich Klocke. That experiment involved 48 university students completing sets of email tasks while being interrupted by instant messages or phone calls.
The number 23 does not appear in that paper.
What the experiment found was more complicated than a simple productivity loss. Participants finished the email task faster when interrupted, after the interruption time itself was removed from the calculation, and made no statistically significant increase in errors. Their emails were slightly shorter. At the same time, they reported more stress, frustration, time pressure and effort during the interrupted conditions.
The authors’ interpretation was that participants compensated by working faster. An interruption did not merely subtract minutes from output; it changed the pace and subjective experience of the work. The study was small, used students rather than employees and involved an artificial email exercise, so it cannot tell us exactly what happens when a programmer, designer or writer loses the thread of a complex project.
Not every diversion has the same cost
The 2005 field study found that interruptions within the same working sphere could sometimes be helpful. A colleague might supply relevant information or move the project forward. Interruptions outside the current context were more likely to force what participants described as a disruptive shift in thinking.
Timing matters too. Leaving a task at a clear stopping point, with useful cues still visible, is different from being pulled away halfway through an unresolved idea. The researchers described a blinking cursor or an open document as a cue that could make resumption easier. Once other windows and subjects displaced those cues, people sometimes struggled even to remember which document they had been using.
This helps explain why a single-purpose medium can feel different from a connected one. In an earlier article on why some readers still prefer physical books, I noted that a paper book functions as a boundary because it does one thing. That observation does not prove print improves attention, but it illustrates how the environment can either limit or multiply possible switches.
The research supports a restrained conclusion. Interruptions can delay the return to a project, often by drawing people through several other tasks, and experimental work suggests that people may compensate with a faster, more stressful pace. It does not establish a universal 23-minute biological reset for the human mind. A stronger answer would require modern, larger studies that measure both real-world task resumption and the quality of concentration after people return.
