AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: A Practical Guide To The Harvard Study On AI And App Switching on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

A Harvard Business School preprint analysed 103 million application events from 1,017 workers at eight companies in India. It found fewer app switches immediately after AI use but more switches overall on days workers used AI; the observational data cannot show that AI caused either pattern.

A Harvard Business School preprint analysing 103 million computer application events from 1,017 workers found that people switched between apps less in the ten uses immediately after using AI, but made about 200 more switches on days they used AI. The study documents an association, not evidence that AI caused workers to focus more or less, and its authors say the paper remains under peer review.

The paper, titled Digital Fragmentation and Generative AI Use Across 103 Million Application Events, is by Sumer S. Vaid and Ashley V. Whillans of Harvard Business School. Its dataset contains 56.9 million app switches across 62,693 person-days, collected from November 2025 to January 2026. The researchers report a mean of 908.3 switches per workday; the median worker made 814, and the middle half ranged from 547 to 1,130.

The paper examines AI use at two scales. In the ten application uses after an AI use, workers used fewer apps and switched less, with each app use lasting about half a second longer on average. But when researchers compared entire days, AI-use days had about 200 more switches, and days with more AI minutes were more fragmented. Those findings are not necessarily contradictory: a short period after AI use can look different from the full day around it.

App stays were short in the dataset: a mean of 27.8 seconds and a median of 3.3 seconds before a switch. Those figures describe app activity, not necessarily uninterrupted thought or task completion. The authors also say day-to-day differences within the same worker accounted for 44.6% of variation in fragmentation, compared with 35.8% between workers and 19.6% between organisations.

At a glance
reportWhen: Preprint posted July 7, 2026, revised A…
The developmentA Harvard Business School preprint examining 103 million workplace application events reports that AI use is associated with different switching patterns over short windows and across whole workdays.

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.

10 app uses before AI median 7 seconds 10 app uses after Whole AI day: about 200 more switches
  1. 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. 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. 3 · The next ten app usesFewer different apps, fewer switches, about half a second longer per app, more predictable. Real, but small.
  4. 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 Use Does Not Equal Less Switching

The findings matter because a widely shared summary implied that AI reduces workplace app switching. The study offers a more mixed picture: switching fell in the short sequence after AI use, while overall switching was higher on AI-use days. Treating the first result as proof that AI improves concentration would omit the other result and go beyond what the research establishes.

For employees and managers, the paper is a reason to be cautious about using app-switch counts as a simple measure of attention or productivity. It did not test whether workers completed tasks faster, produced better work, or felt more focused. Nor does it show that buying an AI tool will reduce interruptions. The researchers’ observation that daily conditions may matter more than stable differences between people or companies suggests teams could pay attention to workload and communication patterns, but that is a practical interpretation, not a tested intervention.

The data also come from workplace monitoring software, making the study relevant to questions about how employers measure digital work. App logs can capture activity at fine resolution, but they do not by themselves explain why a worker changed apps or whether that change helped the work.

Products Worth Considering

DJI Mini 4K Fly More Combo Drone with 4K Camera and 3 Batteries

DJI Mini 4K Fly More Combo Drone with 4K Camera and 3 Batteries

Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight

Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight

As an affiliate, we earn on qualifying purchases.

How the Workplace Data Was Collected

The eight participating organisations were in India and worked across software, accounting, financial services, e-publishing, law and marketing. Records came from Flowace, a commercial workforce-analytics service already used by the companies. Its software recorded when the active application changed, along with app names and other window information. Employers, rather than the researchers, determined which apps counted as productive.

The researchers received anonymised data under a data-use agreement, and the paper says Harvard’s human-subjects committee approved the protocol. The authors report that the research had no external funding and that Flowace had no role in study design, analysis or interpretation. The paper was first posted on arXiv on July 7, 2026, and revised on August 7; its PDF labels it a preprint under review and says findings may change through peer review.

The headline figure of 908.3 refers to the study’s average number of changes between different applications per worker per workday, counting the day from midnight to midnight. It is not a count of every document change, message, or interruption. The paper compares its app-use durations with earlier attention research, but differences in methods mean those comparisons should be read as broad context rather than a directly measured historical decline.

“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 Logs Cannot Establish

The study is observational, so it cannot determine whether AI use led to more daily switching, whether workers turned to AI on days that were already more fragmented, or whether other factors shaped both. The authors explicitly leave open the possibility that both processes occur. The paper also does not measure whether work quality, speed, stress or productivity improved or worsened.

The sample covers workers at eight organisations in India, using employer-provided desktop monitoring software. It does not establish whether the same patterns apply in other countries, among phone users, or in workplaces without similar monitoring. The AI tools recorded included five services, with Claude the smallest in the sample. The extent to which tool mix or job differences affected the results is not settled by the headline figures.

Finally, the paper remains a preprint under review. Its precise estimates and interpretations could change. The detailed dataset reflects monitored application activity, not a complete account of what workers were thinking or doing away from their computers.

Peer Review and Better Measures

The next formal milestone is peer review. Until then, the authors’ posted preprint is the available account of the research, and readers should distinguish its reported associations from causal claims. Further work would need to test whether AI itself changes switching, examine different kinds of work and tools, and measure outcomes such as task completion, quality and worker experience alongside app activity.

For teams responding now, the paper supports caution rather than a promise of a focus benefit. Managers can treat switching as one limited signal, examine how workload varies across the week, and avoid interpreting a high or low app count as a direct measure of performance. The study does not test those workplace practices or show that changing schedules would improve results.

Key Questions

Did the Harvard study find that AI reduces app switching?

Not across the whole workday. It found less switching in the ten app uses after AI use, but about 200 more switches on days when workers used AI. The study does not show that AI caused either pattern.

What does the 908-switch figure mean?

It is the reported mean number of changes between different applications per worker per workday in this dataset. The median was 814 switches; the figure does not count every change of document or every interruption.

Who took part in the research?

The analysis covered 1,017 employees at eight organisations in India, across software, accounting, financial services, e-publishing, law and marketing. Data were collected from November 2025 to January 2026.

Has the paper been peer-reviewed?

No. It was posted on arXiv and is described by its authors as a preprint under review. The authors warn that the findings may change during peer review.

Does the study show that AI improves productivity?

No. The analysis measures application activity and switching. It does not establish effects on work quality, task completion, productivity or focus, and it cannot show that AI caused the observed patterns.

Source: ThorstenMeyerAI.com

You May Also Like

The AI Passed the Crisis Test. Could It Close the Deal?

Five AI models spotted every crisis and refused manipulation, but only two closed the deal. Firmulate’s live experiment asks what happens in your company.

The Significance Of AI Agents Approving Peer Actions

Investigation reveals AI agents sometimes approve each other’s actions without proper authority, raising concerns about control and safety.

Zebkit – Chromium As An Embeddable Library

Zebkit’s GitHub project presents Chromium 157 as a library for native apps, with web, PDF and automation APIs. Its benchmarks are project claims.

AI And The Return-to-Work Story: Inside The Boomerang Data

Surveys report workers returning after AI-related job cuts, but the figures are self-reported and sponsored; the clearest cited reversal involved 45 bank jobs.