Attention Residue: What Switching Tasks Costs and How to Cut It

You finish reviewing a pull request, open a design doc, and read the same paragraph three times. The review is done, yet part of your head is still arguing with a variable name. Researchers call that lingering attention residue, and it may help explain why a day of short, scattered tasks can feel harder than its hours suggest.

Much popular advice repeats the same lines: turn off notifications, stop multitasking, and remember that it takes "23 minutes to refocus". That last claim is not what the research found. This guide returns to the primary papers to show what the experiments measured, where the 23-minute figure comes from, and the intervention with the most direct experimental support we found: a short written plan for returning to the task you are leaving.

If you have one minute: before you leave unfinished work, write down where you stopped, your first action on return, what is still open and what you parked (copy the template). The rest explains why that works, what it does not prove, and how to ration switches with a "switch budget" on days you cannot fully protect.

What Leroy's 2009 experiments showed about attention residue

Sophie Leroy's paper "Why is it so hard to do my work?" (Organizational Behavior and Human Decision Processes, 109(2), 2009, pp. 168–181) is the standard reference for the term. It reports two experiments. Its central claim: to do well on a new task you have to stop thinking about the previous one, and when you can't, performance on the new task suffers. Residue is gauged from performance on that second task and, in 2018, from word associations. The paper is paywalled, so this section rests on its abstract record and Leroy's later work; we could not check what participants did.

Two abstract details are easy to miss. Finishing the first task was not enough for a clean switch, and time pressure while completing it helped people disengage and do better on the next one. The direction matters: pressure to finish helped, while expecting a rushed return hurt, as the 2018 work shows.

Two triggers have named sources. Leroy and Glomb (2018) tested expecting to be rushed when you return; Leroy describes leaving a task unfinished in a 2025 University of Washington Magazine article. The rushed-return finding is the most useful: the same switch costs more or less depending on what you expect when you come back. Experiments like these show that a mechanism exists, not how many minutes a switch costs.

Switch cost, residue, and resumption cost are three different bills

Popular articles blend three literatures. Separating them shows which fix applies. The split is our framing; the studies in each column are cited.

In classic experiments, people alternated between simple rule-based tasks, such as classifying geometric objects or solving arithmetic problems. Rubinstein, Meyer and Evans (2001) found that costs grew with rule complexity and shrank with cues, and Monsell's 2003 review concludes that preparation reduces the cost but does not eliminate it. These are single-trial effects measured in reaction times, and a fix for one bill does not pay the others.

What interrupted people paid in a 2008 email experiment

In Mark, Gudith and Klocke's CHI 2008 experiment, 48 participants (81% German university students, mean age 26) played a human-resources manager answering 12 emails per condition. In one condition nobody interrupted them. In two others a "supervisor" interrupted every two minutes, by phone or instant message, about HR or about something unrelated (how many hot dogs a company party needs). Whether the topic matched made no significant difference to completion time. Compared with no interruptions:

• Time. Excluding time spent on the interruptions, people took 22.8 minutes uninterrupted, against 20.3 (same topic) and 20.6 (different topic).
• Quality. Errors and politeness did not differ. Emails were slightly shorter (29.2 and 30.2 words against 31.5; only the same-topic gap was significant).
• Strain. Ratings of stress, frustration, time pressure and effort were all higher with interruptions (the time-pressure gap for unrelated interruptions was only marginal). On a 20-point scale, stress averaged 6.9 uninterrupted and 9.5 and 9.1 with interruptions.

The authors read this as compensation: people speed up and pay in strain. They add that same-topic interruptions may be perceived as beneficial, yet the actual disruption cost was the same. Our reading: a "quick question about the same project" is not automatically cheap. The limits are real: a simulated office, mostly students, and a task where you can go faster by writing less. Still, it shows why output alone misses the bill.

Where the "23 minutes" figure comes from, and what it measures

The number traces to field observation, not a laboratory measurement of refocusing. Two sources circulate, and they differ:

A working sphere is the 2005 paper's unit of work: a cluster of related events oriented toward a common purpose, such as fixing a client's problem. We could not reconcile the two sets of figures from what is published.

Warning: Both figures measure the delay before people went back to interrupted work, not how long it takes to regain focus. Neither supports "every interruption costs 23 minutes".

The 2005 standard deviation is more than twice the mean, and both figures come from a few dozen people with no comparison group. Retellings drift too: a 2023 UC Irvine repost of a CNN story paraphrases Mark as saying it takes about 25 minutes to "refocus" on a task, but the 2005 paper reports that figure as the delay before work resumed.

What the 2005 data does support is practical:

• Accountable work comes back. Interrupted work the person was accountable for was resumed the same day 82.0% of the time, against 43.8% for work they were helping a colleague with.
• The workspace changes meanwhile. A blinking cursor after your last typed word can reorient you at once; newly opened windows can leave you unsure which document you were editing.
• Many interruptions are self-inflicted. In the 2006 interview Mark said people interrupted themselves about 44% of the time; the 2005 paper's own table puts it nearer 52%, so read it as roughly half. Graduated friction on distracting sites targets one slice of those.

The ready-to-resume plan: evidence, template, and limits

What Leroy and Glomb tested in 2018

Leroy and Glomb (Organization Science, 29(3), 380–397) report four studies, building on research on attention residue, time pressure and implementation intentions (if-then plans). According to the University of Washington summary, one surveyed 202 working professionals and another used word associations to check whether people were still thinking about the first task. In two lab studies with 66 and 44 students, participants wrote a ready-to-resume plan as they left the first task, then reviewed résumés as if making a hiring decision. Plan writers showed less residue, made better decisions and recalled more résumé information. The abstract says the four studies support the prediction that expecting time pressure on your return produces residue and lower performance on the interrupting task.

The university describes the plan as covering where to resume, what challenges are left, and what actions must be postponed. It "needn't be long or elaborate"; a minute is enough.

Note: The plan experiments used lab tasks with students and measured performance on the interrupting task. They did not test whether the plan speeds your return, or programming and writing work; Leroy says there is no reason to believe it wouldn't help on return, but that is untested. We found no source for how long residue lasts, and none that tests the "Back by" line or the switch budget below.

A one-minute resume note you can copy

RESUME NOTE: <task name>
Stopped at:   <exact place: file and function, paragraph, step number>
Next action:  <the first physical step when I come back, verb first>
Still open:   <up to three unresolved problems>
Parked:       <things I noticed and chose to postpone>
Back by:      <a specific slot, e.g. "Thu 09:30", not "later">

The Stopped at, Next action, Still open and Parked lines map onto the university's three elements: where to resume, what is left, what is postponed. Because the 2018 paper builds on if-then planning research, keep Next action concrete (our inference). "Back by" is our addition. Expecting a rushed return drives the effect in the 2018 studies, so a concrete slot may ease that worry, or may simply add a deadline. We found no study that tests it, so book more time than you think you need.

For example, say you are halfway through fixing a billing bug when stand-up starts in two minutes:

RESUME NOTE: invoice rounding bug (#482)
Stopped at:   InvoiceTotal.calc(), line 88, after the discount step
Next action:  Add a failing test for 3 items at 0.335 each
Still open:   Does tax round per line or per invoice? Ask finance
Parked:       Rename calc() to calculate(), separate commit
Back by:      Today 11:00

Outside code, say you are drafting a report section and a colleague needs ten minutes:

RESUME NOTE: Q3 report, supplier section
Stopped at:   paragraph 4, after the sentence on delivery delays
Next action:  Write the one-sentence lead-in to the cost table
Still open:   March or April figures? Check the shared sheet
Parked:       Tidy footnote formatting
Back by:      Today 15:00

The procedure:

1. Check that you are coming back soon. If not, write a ticket comment instead.
2. Write the note before you leave, capped at 60–90 seconds, where your eyes land on return: a comment at the top of the file, a ticket comment, or a pinned note on the page you were reading (the tab note-taking workflow covers the browser case).
3. Close everything except the one file or tab the next action needs; the 2005 paper notes that newly opened windows can make it hard to recall which document you were working on.
4. Switch, and don't reopen the note until you are back.

A small experiment supports preparing before you leave. In Trafton and colleagues (2003), 17 lab employees (8 warned, 9 not) did a tank resource-allocation task and were interrupted for about 30 seconds. With an 8-second lag between alert and interruption, people prepared on about 13.5 of 30 interruptions (judged from think-aloud reports) against 3.1 with no lag, and resumed more quickly. The no-lag group sped up with practice, though, and the preparation was mental rehearsal, not a written note, so treat it as a hint.

Using the note when you return (our sequence, untested)

1. Read the note once before opening anything else.
2. Do the Next action before checking messages. In the 2005 data people worked on 2.26 other things, on average, before resuming; starting with the note skips that detour.
3. Delete or update the note so a stale one does not mislead you later.

Why a plan may quiet the mind, and why not to lean on Zeigarnik

Masicampo and Baumeister (2011) found, across several studies, that unfulfilled goals produced intrusive thoughts, heightened accessibility of goal-related words and poorer performance on an unrelated task, and that making a specific plan for the goal eliminated those effects. The goal did not have to be completed.

The popular claim that unfinished tasks are remembered better (the Zeigarnik effect, from a 1927 study) is on shakier ground. A 2025 meta-analysis by Ghibellini and Meier found no memory advantage for unfinished tasks, though people do tend to resume interrupted ones. Build your case on intrusion and residue, where plans have evidence.

Planning a day with unavoidable switches: the switch budget

A resume note lowers the price of each switch; a budget limits how many you buy. The method is ours, not a published protocol, and the weights are judgement calls.

Score each departure from unfinished work

• 1 point: planned switch, resume note written.
• 2 points: planned switch, no note.
• 3 points: unplanned interruption (someone or something else chose the moment).
• Add 1 if you expect to be rushed when you return.

Treat a cap of 10 as a placeholder, not a research finding. After a week, compare each day's total with how the day felt and move the cap to the score where days start to leave you drained.

A worked example: a developer on the support rota

This is hypothetical; the "after" column changes only what the person can control or negotiate.

The new plan lands at 8, two points under the cap. That leaves room for one more planned switch with a note, but not for another unplanned interruption (8 + 3 = 11). It also needs a team agreement: "urgent" means phone, and everything else waits for a window.

If you are over the cap, work down what you control: write notes for planned switches (2 points becomes 1), move deadline work away from rushed returns (saves 1), and turn unplanned pings into scheduled windows, so several 3-point pings become one 1-point switch (this one needs a team agreement).

Which switches deserve a note

Cut the number of switches, not just their price

• Batch the streams. In Kushlev and Dunn (2015), 124 adults spent one week limiting email checks to three a day and another checking freely; daily stress was lower in the limited week. That measured stress, not residue, and people still averaged 4.7 self-reported checks a day against 12.5 unlimited.
• Remove unplanned entry points. Many unplanned switches start as notifications; the 30-minute notification audit walks through the sources.
• Protect demanding work. Leroy's advice in the 2025 article: work offline, take breaks between tasks and before meetings, do complicated work early, and write things down.

Where to start this week

• Monday: tally every departure from unfinished work with the 1, 2, 3 scoring. Paper is fine.
• Tuesday and Wednesday: write a resume note for every planned departure.
• Thursday: move one stream (chat, email, review requests) into two fixed windows.
• Friday: compare totals with how each day felt, adjust your cap, and add the numbers to your weekly review.

Any timer or notes app that makes you type an intention can hold your Next action line, and so can a plain text file.

Frequently asked questions

Do I still need a note if I finish the task before switching?

Finishing helps, but Leroy's 2009 abstract reports it was not enough on its own. The plan studies tested unfinished work, so a note for finished work is a reasonable extension, not a proven fix.

How long does attention residue last?

None of the sources we could access gives a duration (for Leroy's papers we had abstracts and a university summary, not the full text). The experiments measured performance on the next task, and the 23- and 25-minute figures measure the delay before returning to interrupted work. Treat any "residue lasts N minutes" claim as unsourced.

What if I am the one interrupting someone?

In Leroy and Glomb's studies residue lowered performance on the interrupting task, so a colleague you pull away may give you a weaker answer than usual (our reading). Mark's 2008 experiment found the cost was the same whether or not the question was on topic, and in 2006 she observed that people who sit together can hear natural breaking points in a colleague's work. Without that cue, send a message that can wait (topic, deadline, "reply at a stopping point") and keep calls for what cannot wait (our suggestion).

Is the average attention span really 47 seconds?

A 2023 CNN story, reposted by UC Irvine, quotes Gloria Mark, author of the book Attention Span, saying that average attention on a screen was about 2½ minutes in 2004, about 75 seconds some years later, and 47 seconds now. That is how long people stay on one screen before switching, not their capacity to concentrate, and the excerpt gives no method.

Sources and further reading

• Why is it so hard to do my work? The challenge of attention residue when switching between work tasks — Sophie Leroy, Organizational Behavior and Human Decision Processes, 2009. The original residue experiments.
• Tasks Interrupted: How Anticipating Time Pressure on Resumption of an Interrupted Task Causes Attention Residue and Low Performance on Interrupting Tasks and How a "Ready-to-Resume" Plan Mitigates the Effects — Sophie Leroy and Theresa M. Glomb, Organization Science, 2018. The ready-to-resume plan.
• Task interrupted: A plan for returning helps you move on — University of Washington News, 2018. Sample sizes and plan contents.
• The Cost of Interrupted Work: More Speed and Stress — Gloria Mark, Daniela Gudith and Ulrich Klocke, CHI 2008. The email experiment.
• No Task Left Behind? Examining the Nature of Fragmented Work — Gloria Mark, Victor M. Gonzalez and Justin Harris, CHI 2005. The 25:26 figure and the central versus peripheral split.
• Too Many Interruptions at Work? — Jennifer Robison interviewing Gloria Mark, Gallup Business Journal, 2006. The 23:15, 81.9% and 44% figures.
• Consider it done! Plan making can eliminate the cognitive effects of unfulfilled goals — E. J. Masicampo and Roy F. Baumeister, Journal of Personality and Social Psychology, 2011. Plans reduce intrusion.
• Interruption, recall and resumption: a meta-analysis of the Zeigarnik and Ovsiankina effects — Romain Ghibellini and Beat Meier, Humanities and Social Sciences Communications, 2025. No pooled memory advantage.