🔍 Read the full analysis: How The Harvard Study Frames AI's Effect On App Switching on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
A Harvard Business School preprint analyzing 103 million application events found that workers switched apps less in the ten uses after an AI interaction, but made about 200 more switches on days they used AI. The observational study cannot establish whether AI caused the daily increase, and it did not measure work quality or productivity.
A Harvard Business School preprint analyzing 103 million application events from 1,017 office workers found a split pattern in how AI use related to app switching: workers switched less in the ten app uses after an AI interaction, but made about 200 more switches on days they used AI. The study is observational, not an experiment, so it cannot show that AI caused either pattern.
The paper, “Digital Fragmentation and Generative AI Use Across 103 Million Application Events,” was written by Sumer S. Vaid and Ashley V. Whillans of Harvard Business School. It analyzes second-by-second records collected from November 2025 through January 2026 at eight organizations in India, covering software, accounting, financial services, e-publishing, law and marketing. The authors report 103 million application events and 62,693 person-days of data.
Across the observed workdays, employees switched applications a mean of 908.3 times per day; the median was 814, and the middle half of employees ranged from 547 to 1,130 switches. An app remained active for a mean of 27.8 seconds before a switch, while the median duration was 3.3 seconds. These measures describe application changes, not necessarily changes in task or attention.
The AI findings depend on the time scale. In the ten app uses after an AI use, workers used fewer apps, switched less and stayed about half a second longer in each app. Across the full day, however, AI-use days averaged about 200 more switches, and days with more AI minutes were more fragmented. The authors say the data cannot determine whether AI use increased fragmentation, whether workers used AI on already fragmented days, or whether both occurred.
App switching and AI · one moment
What happens around one AI query?
The study zoomed in on every AI use: the ten app uses before it, and the ten after. Up close, work calms down a little. Zoom out to the whole day, and the picture changes.
- 1 · A workday of switchingWorkers changed apps 908 times a day on average. One app held their focus for 27.8 seconds on average, 3.3 seconds at the median.
- 2 · AI comes in at a busy momentPeople used AI a median of 13 times a day, for a median of 7 seconds. Fragmentation peaked in the hour of AI use.
- 3 · The next ten app usesFewer different apps, fewer switches, about half a second longer per app, more predictable. Real, but small.
- 4 · Zoom out: the whole dayOn days with AI use, the same people made about 200 more switches. Whether AI calms work or busy moments bring AI: unknown.
Source: Sumer S. Vaid, Ashley V. Whillans (Harvard Business School), “Digital Fragmentation and Generative AI Use Across 103 Million Application Events”, arXiv:2607.06681v2, preprint under review. Blocks are schematic, not to scale.
Graphic: Powered by Thorsten Meyer AI
AI Looks Different Across the Workday
The results caution against treating a short-term pattern as evidence that AI improves focus throughout the day. A worker may settle into a narrower sequence of apps immediately after using an AI tool while still experiencing more overall switching on a day when that tool is used. The two findings can coexist because they describe different time windows.
For employers weighing AI tools as a way to reduce interruptions, the paper offers no evidence of a causal productivity gain. It measures application activity, not whether employees completed more work, produced higher-quality results, or felt less interrupted. Its data also indicate that fragmentation varied more from day to day within the same worker than between organizations: 44.6% of the variation was within-person day-to-day, compared with 35.8% between people and 19.6% between organizations. That points to workload and daily conditions as relevant questions, though the study does not establish what drives those swings.
The findings may help managers frame better questions about communication load, schedules and tool use. They do not, on their own, justify buying AI software as a focus intervention or using app-switch counts as a standalone measure of employee performance.
Products Worth Considering

Logitech MX Mechanical Mini for Mac Wireless Keyboard, Space Gray

Bose QuietComfort Wireless Bluetooth Noise-Cancelling Headphones, Cypress Green

Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight
As an affiliate, we earn on qualifying purchases.
How the Study Collected Activity Data
The researchers analyzed data supplied through Flowace, a commercial workforce-analytics service used by the eight participating organizations. Its software records when the active application changes, along with app and window information. Employers determined which apps counted as “productive.” Researchers received anonymized records under a data-use agreement, and Harvard’s human-subjects committee approved the protocol.
The sample included 1,017 employees, with a median of 58 workdays per person. The paper says the research had no external funding and that Flowace had no role in its design, analysis or interpretation. One author disclosed equity in an unrelated company that uses AI to create enterprise digital twins.
The paper was first posted on arXiv on July 7, 2026, and revised on August 7. It labels itself a preprint under review and warns that its findings may change during peer review. A headline circulated on November 18 emphasized the reported 908 daily switches and a reduction after AI use; the latter leaves out the higher switching observed across AI-use days.
“The manuscript is currently under review, and the findings may change based on the peer review process.”
— Vaid and Whillans, in the preprint
What the App Counts Cannot Show
The study does not establish whether AI caused the increase in daily app switches or the lower switching immediately afterward. It also does not measure productivity, work quality, task completion or employee well-being, so app activity cannot be taken as a direct measure of whether work improved or worsened.
The sample covers eight organizations in India and uses employer-provided monitoring data; it does not establish whether the same patterns apply in other countries, workplaces or on phones. The authors’ app-switch measure records a change in active application, not necessarily a change in the employee’s task. The preprint remains under review, and the paper says its results may change.
Peer Review and Broader Evidence
The next formal milestone is peer review; the supplied material gives no publication date or schedule. Readers should check for a revised version or a peer-reviewed paper before treating the estimates as settled.
Further research would need to test whether AI use itself changes switching, rather than simply coinciding with days that differ in workload or communication demands. It would also need to measure work outcomes alongside app activity and examine whether the findings hold in other regions and types of organization. Until then, the study supports a limited conclusion: AI use was linked to different switching patterns at different points in the day, with cause and effect unresolved.
Key Questions
Did the study find that AI reduces app switching?
Only in a limited window: workers switched less in the ten application uses after an AI use. Across the whole day, AI-use days had about 200 more switches, so the paper does not support a simple claim that AI reduces daily switching.
Did AI cause workers to switch apps more?
The study cannot answer that. It observed app activity rather than assigning workers to use or not use AI, so the higher switching on AI-use days is an association, not proof of cause.
Where did the data come from?
Records came from Flowace workforce-analytics software at eight organizations in India. The paper covers 1,017 employees from November 2025 to January 2026.
Is the Harvard study peer-reviewed?
No. The paper was posted as a preprint and says it is under peer review; its authors warn that findings may change.
Does 908 switches a day mean workers changed tasks 908 times?
No. The reported mean counts changes in the active application, not verified changes in task or attention. The study does not measure work quality or productivity.
Source: ThorstenMeyerAI.com
