Employees who use artificial intelligence tools such as ChatGPT at work may be viewed as less trustworthy by their supervisors, potentially affecting performance assessments and career progression, according to a six-study research paper published in _The Leadership Quarterly_.
The researchers conducted five incentivised economic experiments and one field study involving academic supervisors to examine how employees’ use of large language models (LLMs) influences managerial trust.
In one experiment, supervisors were asked whether they would accept an employee’s account of the time taken to complete a task or independently verify the information. The findings showed that employees who completed the task independently received supervisors’ trust 68 per cent of the time. When employees used AI-generated output, the likelihood fell to 24 per cent.
AI-assisted work, however, was viewed more favourably than work copied from another person.
The researchers noted that the experiment focused on supervisors’ willingness to trust employees’ accounts rather than the quality of the final output. This suggests that the knowledge of AI involvement itself influenced perceptions of employee trustworthiness.
The consequences extended beyond immediate trust.
Across the studies, lower supervisory trust was associated with poorer perceived performance, lower assessments of organisational citizenship behaviour and reduced perceptions of employees’ promotability.
The research also found that organisational policies played a significant role in shaping supervisors’ reactions. When companies explicitly encouraged employees to use LLMs, the negative effect on supervisory trust was no longer statistically significant.
However, the trust penalty remained when organisations prohibited AI use or had no clearly stated policy on its adoption.
The researchers also examined whether task importance or supervisors’ own experience with AI influenced their responses. Neither consistently reduced the negative effect.
A field study involving academic supervisors evaluating graduate students’ work produced similar findings. The broader the students’ use of LLMs, the lower the supervisors’ reported trust in them.
The findings raise questions for employers that are encouraging AI adoption while leaving employees uncertain about how its use will affect their professional standing.
Employees may feel compelled to conceal their use of AI tools if they believe admitting to it could damage their credibility or career prospects. This can contribute to what is commonly described as ‘shadow AI’, where workers use AI tools without formally disclosing that they are doing so.
For organisations, the findings suggest that introducing AI tools alone may not be sufficient. Clear guidelines on acceptable use, alongside consistent communication with managers, could help employees understand when and how AI-assisted work is appropriate.
The study does not establish that every employee using ChatGPT will face lower trust or reduced promotion prospects. Instead, it highlights how managerial perceptions of AI use can influence workplace evaluations, and how explicit organisational support can change those perceptions.
