Artificial intelligence tools are becoming routine in offices, promising quicker drafts, faster research and a way to offload repetitive tasks. But a new study suggests that the social consequences of using those tools may not track neatly with their productivity benefits. When supervisors know an employee used a large language model such as ChatGPT, they may be less inclined to trust that employee—even when the technology is not prohibited and even when the quality of the work itself is not under review.
The research, led by Samuel Meyers of the Hebrew University of Jerusalem with Yochanan Bigman of the Hebrew University and Roman Briker of WHU—Otto Beisheim School of Management, was published in The Leadership Quarterly. Across six studies, the researchers examined how supervisors respond to employees’ use of large language models and how those reactions could shape workplace evaluations.
Trust changed when the method was known
A central finding was that managers’ judgments were affected by the knowledge that AI had been used, not simply by the finished result. In one experiment, people acting as supervisors decided whether to trust an employee’s report about the time required to complete a task or to verify it themselves. Participants were told whether the employee had completed the work independently or with AI assistance.
Employees described as having completed the work themselves had a 68% likelihood of receiving their supervisor’s trust in that scenario. For employees described as having used AI, the figure was 24%. The researchers emphasized that supervisors were not assessing the quality of the completed work in this experiment. The difference instead points to the perceived significance of the production process.
That distinction matters as organizations try to establish practical rules for generative AI. A company might regard AI as an approved efficiency tool, while individual managers may still read reliance on it as a signal about an employee’s effort, judgment or dependability. The study does not establish that every manager will respond in this way, but it does identify a potential gap between a tool’s formal acceptability and its interpersonal consequences.
Professional judgments can follow
The research linked lower supervisory trust with broader professional assessments. In additional experiments, lower trust was associated with poorer evaluations of employee performance, weaker perceptions of an employee’s contribution to the organization and lower perceived suitability for promotion.
Those findings make AI disclosure more complicated than a simple question of transparency. Workers may want to be open about the tools used in their work, particularly where accuracy, confidentiality or organizational policy are involved. Yet the study suggests that disclosure can trigger a negative interpretation before a manager considers the result on its merits. The risk is especially relevant for work where contribution and ownership are difficult to observe directly.
The comparison with copying from another person adds an important boundary. AI-assisted work was viewed less favorably than independently completed work, but more favorably than plagiarism. In other words, participants did not treat AI use as equivalent to copying someone else’s work. Still, it did carry a distinct trust penalty in the research scenarios.
An organizational challenge, not only an employee choice
Generative AI is often discussed as a personal productivity decision: employees choose whether to use it, and managers decide whether the output meets expectations. This study frames the issue more broadly. If managers and employees bring different assumptions to AI-assisted work, then the resulting tension can affect trust, feedback and advancement.
Clear workplace expectations may therefore need to cover more than whether AI is permitted. They may also need to address what kinds of assistance are appropriate, when it should be disclosed and how managers should evaluate work that combines human and machine input. A policy that simply says employees may use AI does not by itself resolve how supervisors interpret that use.
For employees, the study is not a reason to assume that any use of an AI tool will damage a career. Its evidence comes from experimental studies of supervisory judgments, and workplaces differ in their norms, tasks and oversight. But it is a warning that the adoption of AI can change workplace relationships as well as workflows.
As Bigman put it, AI is changing not only how people work, but also how they judge the people they work with. The next phase of workplace AI adoption may depend as much on building shared expectations around trust and accountability as on choosing the most capable tools.
This article is based on reporting by Phys.org. Read the original article.
Originally published on phys.org
