Nearly Half of Required AI Users Resist It—Wellness May Explain Why

At its HR Symposium/Xpo in London on October 8, 2026, Gartner reported findings from a first-quarter survey of 11,838 employees: 47% of those expected to use AI at work resisted adopting it, while fewer than 35% of employees reported good overall wellness; senior principal analyst Nicole Paripurana said AI could support “the human side of work.” The results put a question beside the proposal to use AI to detect unhealthy work patterns: would employees experience it as help or as closer observation?

The figures describe different groups within the survey. The published findings do not show whether employees reporting poor wellness are the same people resisting AI, so they do not establish that one causes the other. They do, however, make employee experience central to the adoption question: a tool that promises an organizational benefit may be harder to accept if using it adds pressure or reduces control over work.

What the resistance figure measures

The 47% figure applies to employees expected to use AI in their work. It is a measure of reported resistance within that group, not a measure of how many employees have stopped using a particular product. The release gives no breakdown by tool, occupation or prior use, leaving open whether reluctance concerns the technology itself, the way it is introduced or the work that follows.

An October 10 account of the London findings describes an asymmetry in employee attitudes: many see how AI could benefit their organization but are less convinced it will benefit them personally. That distinction matters when a faster task leads to more work to review or a higher output target. Those are possible reasons an employee might resist adoption, rather than explanations measured by this survey.

Wellness may be part of that experience, but the connection needs to be kept at the level the evidence allows. An employee already carrying an excessive workload could welcome a tool that helps reduce it. The same employee could reasonably question a tool that records signs of strain while leaving the workload intact. The difference lies in what changes for the worker after the signal appears.

What AI wellness support would actually do

The proposed uses begin with patterns in ordinary work activity. Frequent switching between tasks, changes in communication tone and signs of exclusion or disengagement could be treated as early indications that someone needs support. Other signals could point to excessive workloads, burnout risks, toxic behavior, bias or violations of work boundaries. These are proposed ways to identify concerns, not reported results from deployed wellness systems.

The proposal also includes tailoring support to individual preferences, such as tolerance for meetings, preferred forms of connection, an opt-in for stress signals and customized well-being nudges. A reminder to protect uninterrupted time may be useful when the employee can decide whether to act on it. An inferred account of that employee’s stress becomes a different intervention if it is routed to a manager or retained for later assessment.

Neither a change in tone nor a quiet period in a communication channel explains a person’s health, motivation or performance on its own. Context matters: the employee may be doing focused work, dealing with a temporary change in duties or communicating through another channel. If a system treats such observations as conclusions, its apparent precision could give a weak inference more authority than it deserves.

When a support signal becomes surveillance

The same data that could reveal overload could also expose details of an employee’s daily pace and interactions. The International Labour Organization’s analysis identifies intrusive surveillance, work intensification, reduced autonomy and concerns about privacy and data use as risks to psychosocial working conditions. That makes the destination and use of a wellness signal as consequential as the system’s ability to detect it.

Meaningful consent would require clarity about which activity is analyzed, who can see an inference, how long it is kept and whether an employee can decline or withdraw without consequences for their work. Those are criteria for assessing a proposed system, not safeguards established in the survey release. An opt-in loses much of its value if declining it affects access to work or if personal signals later feed into performance decisions.

Escalation also needs a boundary. A possible sign of distress could lead to a private offer of human support, while a pattern of excessive workload could prompt a review of staffing or deadlines. Automatically passing an individual stress score to a line manager would expose the employee to a different kind of scrutiny. Those choices cannot be settled by the accuracy of a model alone; they determine whether workers have agency over information drawn from their behavior.

The employer still owns the workload

An alert about overwork is useful only if someone with authority can change the conditions behind it. A nudge to take a break cannot reduce an assigned workload, move a deadline or resolve conflicting priorities. Managers and employers remain responsible for those decisions even when AI helps them notice a pattern sooner.

That responsibility bears directly on adoption. Workers asked to use AI will encounter both its immediate effects on their tasks and any wider system that interprets their behavior. If an employer uses wellness signals to adjust work and offers support under clear limits, the tool may address a concern employees can recognize. If the signals become another measure of individual performance, the promised support may deepen the distrust the employer hoped to overcome.

Also read:

• TikTok’s Safety Reset Allegedly Did Nothing for Thousands of Young Users