Here is the arithmetic. Gloria Mark of UC Irvine has said that in her research, interrupted work was resumed on average 23 minutes and 15 seconds later, usually after about two other tasks in between. In an earlier observational study of information workers, she and Victor González found that people spent about three minutes on a task before switching, counting switches they made themselves as well as outside interruptions.
Put those two numbers side by side. If getting back to interrupted work takes that long and switching happens that often, a lot of the day is spent partway back to something. Often you never fully recover. You work in a state of continuous partial attention, accumulating cognitive debt that compounds throughout the day.
By 3pm, your brain does not just feel broken. By the numbers, it is operating in a deficit it cannot clear.
What Happens Neurologically
A notification is not a neutral event. Even a single ping triggers a specific neurological sequence that has real cognitive costs.
Step 1: Threat assessment
Your brain’s attentional system registers the notification. Before you consciously process it, your amygdala performs a rapid threat assessment. Is this urgent? Does this require action? This happens in milliseconds, but it pulls resources from whatever you were doing.
Step 2: Context switch
If you check the notification, your brain must swap context. The current task’s working memory contents are displaced by the new input. This is not instant. Context loading takes measurable time and cognitive effort.
Step 3: Cognitive load transfer
The notification introduces new information that requires processing. A question needs an answer. A message needs evaluation. An update needs integration into your mental model of a project. Each of these is a decision or a set of decisions.
Step 4: Processing and response
If you respond, the effort is obvious. But even if you do not respond—even if you just read the notification and return to your task—the content persists in working memory. Sophie Leroy at the University of Washington calls this attention residue.
Step 5: Attempted restoration
You return to your original task. But the original context must be reloaded. Where were you? What were you about to do? What was the logic chain you were following? This reconstruction takes time, and it is imperfect.
Attention Residue (Sophie Leroy, University of Washington)
When you switch from Task A to Task B, part of your attention remains on Task A. This “residue” reduces performance on Task B. The effect is stronger when Task A was not completed before switching—which is precisely what happens with notification interruptions. Working memory can hold only about four items at once (Cowan, 2001).
The Hidden Costs
The 23-minute figure is the headline number, but the less visible costs may be more significant.
Decision fatigue acceleration
Each notification forces a cascade of micro-decisions: Should I check this? Is it urgent? Should I respond now or later? What does this mean for my current task? These decisions draw from the same limited pool of cognitive resources you use for your actual work. Notifications do not just interrupt your task—they consume the resources you need to resume it.
Attention training degradation
The brain adapts to its environment. A notification-heavy environment trains the brain to expect interruptions, which makes sustained attention harder even during quiet periods. Research on heavy media multitaskers shows measurably reduced ability to filter irrelevant information and maintain focus—effects that persist outside the multitasking context.
Recovery time inflation
Recovery time is not constant. The 23-minute figure is an average, and we suspect, without a study to cite, that recovery takes longer when you are already tired. The tenth interruption of the day is harder to recover from than the first. By afternoon, recovery from a single notification may take significantly longer than the average suggests, because the baseline has shifted.
This assessment tool has been retired
It was part of our original decision-load research. Our current work applies the same governance thinking to how organizations run autonomous AI agents.
Take the Governance Stress TestThe Decision Load Connection
Each notification is not just an interruption. It is a decision point—often several decision points.
A single Slack message might require:
- Deciding whether to read it now or later
- Evaluating the content for urgency
- Deciding whether to respond, and if so, how
- Assessing impact on current priorities
- Deciding when to follow up if deferred
Five decisions from one notification. Multiply by the number of notifications per hour—for many people, dozens a day—and the decision volume is substantial. These are not complex decisions individually. But they accumulate. And the cognitive cost of decisions is cumulative, not isolated.
This is why the “just ignore notifications” advice does not work. Even the decision to ignore a notification is a decision. Even the awareness that notifications exist in the background occupies cognitive bandwidth. The problem is structural, not behavioral.
What Reduces the Damage
The research points toward several approaches that measurably reduce notification-related cognitive depletion.
Batched attention
Processing notifications in scheduled batches rather than in real-time reduces the total number of context switches. Research found that limiting email checks to three times daily significantly reduced stress compared to unlimited checking (Kushlev & Dunn, 2015).
Context preservation
Before checking a notification, writing a brief note about where you are in your current task (even just a few words) significantly reduces attention residue when you return. Sophie Leroy's research on "ready-to-resume" plans found that this simple step reduced attention residue and protected performance on the interrupting task (Leroy & Glomb, 2018).
Notification gradient
Not all notifications are equally costly. A notification about a direct message to you has a different urgency profile than a channel update. Most notification systems treat all alerts identically. Configuring notifications into tiers—immediate, batched, and daily summary—reduces total interruption volume without missing genuinely urgent items.
Measurement
Tracking your cognitive state at regular intervals reveals patterns that are invisible in the moment. Many people discover that their heaviest notification periods correlate precisely with their afternoon cognitive crashes—a connection that feels obvious in data but is difficult to perceive in real-time.
Recovery protection
The 23-minute figure describes how long people took to get back to interrupted work, not a fixed recovery period, and further interruptions only lengthen it. Protecting even short blocks of uninterrupted time (30 to 60 minutes) allows cognitive debt to partially clear. The key is that these blocks must be genuinely uninterrupted—not “mostly” uninterrupted with occasional quick checks.
The Structural Problem
Individual notification management helps, but it does not address the structural issue: modern knowledge work environments are designed around real-time communication in a way that conflicts with how human cognition works.
Slack, Teams, email, and their equivalents assume that faster communication is better communication. The research suggests otherwise. Faster communication means more interruptions, which means more cognitive debt, which means lower-quality work on the things that actually matter.
The fix is not willpower. It is not about being better at ignoring notifications. It is about understanding the actual cost of each ping—not just the seconds it takes to read, but the minutes of recovery, the decisions it forces, and the cumulative effect on your capacity to do meaningful work.
Your brain is not broken. It is operating exactly as designed—in an environment that asks it to do something it cannot do: maintain deep focus while processing a continuous stream of interruptions that can each take many minutes to come back from, in a world that switches you every few.
This assessment tool has been retired
It was part of our original decision-load research. Our current work applies the same governance thinking to how organizations run autonomous AI agents.
Take the Governance Stress TestCorrections, 28 September 2026: The 23-minute figure is Gloria Mark’s average time before interrupted work was resumed, reported in an interview, not a measured time to refocus after each notification (Gallup interview). The “interrupted every 3–6 minutes” figure was replaced with what the study found: about three minutes on a task before switching, including self-initiated switches (González & Mark, CHI 2004). We removed four figures we could not trace to any study: 7.2 active cognitive threads, checking notifications within 2 minutes, 12 to 15 minutes of residue attributed to Carnegie Mellon, and 50 to 80 notifications a day. The Kushlev & Dunn (2015) summary no longer claims unmeasured effects on response quality, and the Leroy & Glomb (2018) result is now described as it was reported, better performance on the interrupting task (Leroy & Glomb, 2018).