ActivTrak Workday Study Finds AI Users Have Longer Days

ActivTrak reports that AI users have workday spans up to 90 minutes longer than nonusers, challenging the promise that faster tasks deliver shorter days. The ActivTrak workday study covers 9,140 knowledge workers across 467 enterprises, using digital activity collected between April 1 and September 25, 2026. Released October 29, the findings connect AI adoption with longer days, without establishing that AI caused the difference.

That distinction matters because the apparent contradiction has several possible explanations. Employees with demanding jobs might adopt AI more often. Faster production might encourage additional assignments. Moving work between assistants and existing applications might also consume time that task demonstrations leave out. Each explanation puts pressure on employers promising efficiency without measuring what happens to the entire working day.

Earlier AI workday research provides an essential counterweight. A randomized experiment involving an assistant integrated into existing applications found reduced email time and less work outside regular hours. The emerging conflict is therefore between a productivity promise and uneven workplace outcomes. AI can save time on an assignment while leaving unresolved whether employees finish earlier, handle more assignments, or absorb additional checking and coordination.

What the ActivTrak Workday Study Actually Found

The latest finding concerns the span of the workday, which should not be confused with continuous labor or verified output.

According to ActivTrak’s workday analysis, AI users’ days spanned 8.4 to 9.1 hours, compared with 7.6 hours for nonusers. Heavy users averaged uninterrupted focus sessions of 37.8 minutes, compared with 48.2 minutes among light users. Heavy users also recorded 18 additional minutes of daily after-hours activity.

Those measurements describe different aspects of work. A longer span does not establish uninterrupted effort throughout that interval. Shorter focus sessions do not establish lower quality. Additional evening activity does not identify whether an employer requested it, an employee chose it, or scheduling flexibility shifted it.

The headline figure also represents an upper comparison within the reported ranges. It should not become a claim that every employee adopting AI adds the same amount of work. Treating an observed difference as a universal penalty would erase precisely the variation that employers need to investigate.

The timing makes the release relevant to enterprise planning. Organizations evaluating AI investments need evidence beyond whether employees have opened an assistant. They need to understand whether its use changes delivery times, staffing needs, error rates, and working boundaries.

Consider the distinction between a faster draft and an earlier departure. A drafting assistant can reduce the effort required to produce an initial document. That leaves open what happens during approval, fact checking, revision, and distribution. The completed assignment remains the relevant unit for assessing a workplace benefit.

The study does not resolve that entire chain. Its contribution is to expose a gap between using AI and having a shorter day. That gap deserves investigation even when the direction of causation remains uncertain.

For employees, the practical implication is straightforward. Feeling faster at isolated tasks does not contradict feeling busier overall. Those experiences measure different things, and an employer’s productivity assessment needs room for both.

For buyers, the release raises a procurement question. A product demonstration typically ends when the assistant returns an answer. A business process ends when someone accepts responsibility for the result. The distance between those endpoints is where an apparent saving needs verification.

Managers Face a Choice About Where Saved Time Goes

A shorter task creates available capacity; management decisions determine whether that capacity becomes relief, additional output, or another obligation.

The immediate pressure falls on managers approving AI deployments and employees expected to demonstrate their value. Buyers need evidence that investment improves business performance. Workers need clarity about whether faster execution changes their responsibilities, deadlines, or availability.

These interests can align. An employee might prefer removing routine drafting so they can concentrate on work requiring judgment. An employer might value that shift more than reducing scheduled hours. The problem begins when each side assumes a different destination for the same saved time.

A shorter workweek is a specific outcome. It requires work to fit within fewer scheduled hours, with acceptable quality and manageable demand. A productivity improvement alone does not guarantee that arrangement. Organizations can instead direct the additional capacity toward an existing backlog.

That choice is not automatically unreasonable. Unfinished work can include useful projects, customer requests, or neglected maintenance. However, leaders should describe the outcome accurately. Delivering more work within the same schedule differs from reducing the schedule, and extending the day creates another outcome entirely.

The ActivTrak AI workload discussion makes this distinction harder to avoid. Employers cannot assess employee benefit solely through the volume of generated material. An increase in drafts might help the author while increasing demands on editors, reviewers, or colleagues responsible for implementation.

A hypothetical product team illustrates the issue. A manager uses AI to prepare a proposal faster and adds further alternatives before sharing it. The team then evaluates more options, checks more assumptions, and discusses more possible changes. The initial task becomes cheaper, while the decision process becomes larger.

Nothing in that example requires the assistant to malfunction. Every draft might be useful. The additional burden arises from expanded ambition and limited review capacity. Faster production changes what appears feasible before the organization changes what it can responsibly approve.

Managers therefore face a problem that extends beyond prompting skill. Someone must decide which additional work deserves attention and which work should remain undone. Without that decision, an assistant can expand the supply of proposals faster than colleagues can evaluate them.

The same issue applies to knowledge management. Organizing existing information can address a different burden from generating more material. A personal knowledge base concerns how information is retained and retrieved. Neither function should receive credit for shorter working hours without evidence that the overall process has improved.

The relevant management question is therefore concrete: what stopped consuming employee time after deployment? If nothing disappeared, leaders should identify the additional value that justifies the new activity. Otherwise, adoption becomes an expectation layered onto an already full schedule.

Faster Tasks Can Expand the Job Around Them

AI’s ability to make work easier can also increase how much work employees attempt.

Research from UC Berkeley Haas offers a grounded account of that possibility. Xingqi Maggie Ye and Aruna Ranganathan conducted an eight-month ethnographic study, using workplace observation and interviews, at a technology company with 200 employees. Ye also conducted more than 40 interviews across functional groups.

Their work intensification research identified broader responsibilities, work entering former breaks, and more simultaneous tasks. Employees sometimes kept AI processes running while handling other obligations. The research describes how work expanded, rather than estimating a universal effect across employers.

The distinction matters for interpreting the ActivTrak workday study. Workplace observation can reveal mechanisms that activity measurements cannot explain by themselves. It does not prove those mechanisms caused the differences in another dataset. It does, however, identify explanations worth testing.

One possibility is that reducing the difficulty of starting work changes the stopping point. An assignment that previously seemed impractical can become worth attempting. A worker who can quickly explore an idea faces a new decision about whether to continue, refine, and share it.

That creates a different experience from simply receiving an extra task from a manager. The additional effort can begin with curiosity or satisfaction. An employee might value the ability to attempt work outside their previous skill range. Enjoyment and workload expansion can therefore occur together.

Aruna Ranganathan’s research overview emphasizes this distinction between measurable production and employees’ lived experience. Her work examines autonomy, cognitive demands, expertise, relationships, and well-being alongside other consequences of workplace AI. Those dimensions help explain why a favorable output measure cannot settle every question about a deployment.

The analytical concern is accumulation. A small optional experiment becomes a useful deliverable. Useful deliverables influence expectations. Expectations then shape what colleagues request and what employees feel responsible for maintaining. The original saving can disappear without a formal decision to expand anyone’s role.

Parallel work introduces another version of that problem. Starting an assistant does not eliminate responsibility for checking its result. If several processes finish while a worker handles another assignment, their outputs create a queue requiring attention.

The resulting burden depends on what those outputs require. Routine material might need little review. A consequential recommendation might require source checking, consultation, and revision. Counting the assistant’s completed responses would treat those situations as similar despite their different human demands.

The reversal is therefore plausible without assuming AI has failed at its immediate task. Better assistance can increase the range of work people undertake. Whether that becomes a benefit depends on workload limits, review capacity, and the value of the additional results.

Integrated Assistants Offer a Different Reference Point

Evidence that some AI users work longer does not erase experimental evidence that other deployments save employee time.

A study by Eleanor Wiske Dillon, Sonia Jaffe, Nicole Immorlica, and Christopher T. Stanton provides an important comparison. Their randomized workplace experiment involved 7,137 knowledge workers across 66 firms. Random selection determined access to a generative AI assistant integrated into applications used for email, meetings, and writing.

In the latter half of the six-month experiment, participating users spent two fewer hours on email each week and reduced work outside regular hours. The researchers did not detect changes in task quantity or composition resulting from individual access. These findings establish a different possible outcome from the longer spans reported in observational data.

The comparison strengthens the need for careful claims. It does not establish that integrated assistants always shorten days or that standalone assistants always extend them. Different samples, methods, applications, and working environments prevent that shortcut.

It does establish that “AI makes employees work longer” is too broad as a universal conclusion. Access can reduce time spent on particular activities under some conditions. An observed association elsewhere cannot invalidate that experimental result.

The distinction between products also deserves attention without becoming a vendor contest. ChatGPT, Claude, and Microsoft Copilot should not be treated as interchangeable workplace deployments. An assistant’s location within a process, available context, and required review affect what employees must do around its output.

This is a comparison of deployment choices, rather than a claim about which brand performs best. The cited evidence does not support a ranking across those products. Buyers need to evaluate the specific version, configuration, and process they intend to use.

A hypothetical writing workflow makes the distinction tangible. If an assistant returns material within the document being edited, the worker can assess it beside the surrounding text. If material arrives elsewhere, the worker must move it, reconcile context, and verify that the final document still reflects the intended sources.

Neither arrangement guarantees success. The example identifies additional steps that a task demonstration might exclude. A fair evaluation needs to count those steps wherever they occur.

The same reasoning applies to assistants that retrieve internal knowledge. Answer generation is only part of the assignment when employees must confirm which document is authoritative. A confident response can still require substantial checking if its supporting context is unclear.

For enterprise buyers, the competitive question becomes whether a deployment removes work across the completed process. More convenient generation is useful, but its value depends on what follows. The experimental evidence gives buyers reason to pursue measurable savings while resisting blanket promises about shorter weeks.

Longer Days Still Do Not Establish Cause and Effect

The ActivTrak workday study identifies a pattern that deserves investigation, while leaving important explanations unresolved.

The most obvious issue is selection. Employees do not necessarily adopt AI at random. Those handling complicated assignments or heavier workloads might have stronger reasons to seek assistance. Their longer days could partly precede their AI use.

Role differences complicate interpretation further. A job centered on writing, coordination, or coding creates different opportunities for using an assistant. Comparing workers across those roles can mix the effect of AI with the demands of the work itself.

Reverse causation is another possibility. An approaching deadline might lead employees to work longer and use AI more frequently. In that situation, the workload helps explain both observations. The assistant might reduce some effort even while its users continue recording longer days.

These are explanations to test, rather than established accounts of this sample. They show why a correlation cannot answer the causal question alone. The public release does not provide a randomized design that settles those alternatives.

Measurement presents another limit. Digital activity reveals aspects of how people interact with applications. It does not directly establish the value of their decisions, the quality of their collaboration, or their level of exhaustion.

A longer day might accompany valuable output, avoidable rework, flexible scheduling, or a mixture of those conditions. Distinguishing them requires information beyond the duration itself. Employee experience and accepted results belong alongside activity measures.

Earlier findings also need careful separation from the latest release. ActivTrak’s annual workplace analysis included a subset of 10,584 users compared before and after AI adoption. It reported increases in measured work categories, including email and messaging. That is a different analysis from the October comparison, and its sample should not be combined with the newer one.

This separation prevents an inflated impression of certainty. Multiple findings can point toward greater activity while measuring different populations and periods. Agreement in direction does not automatically identify a shared cause.

ActivTrak’s commercial position deserves consideration as well. The company sells workforce analytics, giving it an interest in how businesses measure work. That does not invalidate the findings. It makes transparent definitions, independent replication, and access to methodological detail especially useful.

The strongest conclusion is narrower than the most dramatic headline. AI adoption does not reliably demonstrate employee relief. The ActivTrak AI workload findings warrant further investigation, while experimental evidence shows that time savings remain achievable.

That leaves employers with a practical responsibility. They should distinguish additional output from reduced effort when describing results. Employees deserve to know which outcome a deployment is intended to produce and how success will be assessed.

What Would Show That AI Is Giving Time Back

The next convincing evidence will connect accepted work, employee hours, and quality within the same evaluation.

Over the coming months, the first signal to watch is research following comparable workers before and after deployment. Stronger designs would account for differences in roles, workload, and seasonal demand. Randomized access, where practical, would help separate the effect of the assistant from the reasons employees adopt it.

If longer days persist under those conditions, the argument that deployment contributes to extended working time becomes stronger. If the difference largely disappears, selection becomes a more convincing explanation. Either result would improve on treating all AI users as a uniform group.

The second signal is an employer reporting what happened to saved capacity. A claim that employees completed tasks faster leaves the central question unanswered. Evidence should establish whether accepted output increased, schedules contracted, backlogs shrank, or review demands changed.

A shorter scheduled week would directly support the promise of reduced working time. Greater output within stable hours would support a different benefit. Increasing output alongside longer hours would require a clearer account of whether productivity per hour improved.

Quality must remain part of that assessment. A faster draft that needs extensive correction cannot carry the same weight as accepted work. Equally, additional review does not erase a benefit when the finished result improves enough to justify it.

The third signal is evidence tracking workflow changes alongside employee outcomes. Buyers should watch whether removing handoffs, consolidating review, or improving source access changes working hours. The relevant comparison is the completed process before and after redesign.

If those changes reduce evening work while preserving quality, the case for better implementation strengthens. If hours remain extended, attention should shift toward workload expectations and organizational demand. An easier interface cannot determine how much work an employer asks people to complete.

These signals should appear together in credible reporting. Hours alone leave output unclear. Output alone leaves employee effort unclear. Usage alone says little about either.

For knowledge workers, AI workday research is most useful when it explains those relationships rather than issuing a universal verdict. A tool can be helpful, enjoyable, and associated with an expanding job. A deployment can also produce real time savings without delivering a shorter workweek.

The ActivTrak workday study puts that distinction in focus. Its finding challenges automatic promises of relief, while leaving room for better designs and different outcomes. The question now is whether organizations will measure where the saved effort actually goes.

Before accepting another productivity claim, look for evidence about the whole assignment and the whole working day. Ask what work disappeared, what additional work arrived, and who checked the result. Compare those answers with accepted output, errors, and employee experience. A credible account should explain whether people gained time, delivered more value, or simply remained available longer. It should also distinguish an initial experiment from a sustained change in working conditions. Use those questions when evaluating the next assistant rollout or workplace study. If an employer says AI is giving time back, what evidence shows where that time went?