The 23-minute number gets quoted constantly. It's become a talking point for productivity blogs and workplace wellness decks. The actual research it comes from is more unsettling — and more specific — than any of those uses suggest.
Who Gloria Mark is and what she actually measured
Gloria Mark is a professor of informatics at UC Irvine who has spent over two decades studying how people interact with digital tools at work. Starting in the early 2000s, she and her team did something remarkably low-tech: they followed knowledge workers around offices with clipboards and stopwatches, observing and recording every task switch, every interruption, and how long it took people to return to their original work.
The 2004 study — the one that produced the 23-minute figure — involved shadowing 24 workers across two tech companies for three days each, recording 4,517 task-switching events. The average time to return to a task after an interruption was 23 minutes and 15 seconds. This is the number that got quoted. It is not the most important number in the study.
The more important finding: people rarely return to the original task at all
After being interrupted, participants in Mark's study redirected to an average of 2.26 other tasks before returning to the original task — if they returned at all. In approximately 40% of cases, people never returned to the interrupted task during the observed period. They had been interrupted into a new priority, which led to another new priority, and the original thread was simply dropped.
This is not about recovery time. This is about task survival. The question isn't how long it takes you to refocus. The question is whether you get back to the task at all. Interruptions don't just slow you down. They terminate threads of work that never get resumed.
The self-interruption problem (which is worse)
By the time Mark published a follow-up study in 2012, something had changed significantly. External interruptions — a colleague stopping by, a manager calling — had declined as a proportion of total interruptions. Self-interruptions had increased sharply. People were now interrupting themselves more than external sources were interrupting them.
The mechanism: constant connectivity had internalized the interruption reflex. People check email not because an email arrived, but because they're habituated to checking. They switch tabs not because something demands attention, but because the discomfort of sustained attention has a lower threshold than it used to. The external environment became less disruptive precisely as people learned to disrupt themselves.
Mark's 2014 research, tracking employees with biometric sensors, found a direct correlation between self-interruption frequency and elevated cortisol levels — the stress hormone. People were stressing themselves out by interrupting themselves, not by being interrupted.
What Rubinstein, Meyer, and Evans found about task-switching
Gloria Mark's work is observational. In 2001, Joshua Rubinstein, David Meyer, and Jeffrey Evans published a controlled study in the Journal of Experimental Psychology: Human Perception and Performance that measured switching costs in controlled conditions — specifically, the time lost when the brain has to reconfigure itself to operate on a different task.
Their finding: switching between tasks always imposes a cost, even when the switch is voluntary and expected. The cost has two components. The first is "goal shifting" — deciding to switch — which takes a measurable amount of time. The second is "rule activation" — loading the rules for the new task into working memory while deactivating the rules for the previous task. This second step takes longer for complex tasks and longer when the tasks are cognitively similar (because they use overlapping mental resources).
The practical implication: switching from writing to checking a related document is not costless, even if both feel like the same kind of mental work. The similarity is part of the cost, not a mitigation of it.
How your app stack creates structural switching costs
Most knowledge workers operate across a minimum of five to ten apps in the course of a single project: a task manager, a notes app, a communication tool, a document editor, a reference manager, a browser with multiple tabs. Each transition between these tools is a task switch, even if the underlying cognitive work is continuous.
The switching cost doesn't care whether you intended to switch. It doesn't care whether the switch took one second. The rule-activation and goal-shifting costs are incurred regardless of how smooth the transition feels. The smoothness is an illusion produced by familiarity with the switching routine itself — not evidence that no cost is being paid.
For project work specifically, the highest-cost switch is between notes and tasks on the same project — the case where you need to check a note to understand why a task exists, or need to create a task from something you wrote in a note. These switches involve closely related cognitive content, which means the deactivation cost is high. You have to partially unload what you were thinking to pick up the adjacent thing.
The design implication
The cognitive science points in a consistent direction: every app boundary in your workflow is a potential task-switch point, and task switches are not free. The consequences are not just wasted seconds but dropped threads, elevated stress, and a systematic bias toward whatever is most recently interrupted — which is usually the least important thing, the most recently arrived thing.
Reducing app boundaries for related work reduces structural switching costs. This is not about minimalism as an aesthetic or about simplicity as a value. It is about the cognitive overhead created by tool fragmentation — overhead that Gloria Mark and others have measured with stopwatches, cortisol tests, and controlled experiments over two decades.
The 23-minute recovery number is striking. But the finding that workers interrupt themselves more than they get interrupted, and that those self-interruptions correlate with measurable physiological stress, is the one that should change how you think about your setup.