Bosses Trust Workers Less When They Use AI, Study Finds

AI Use May Cost Workers Credibility With Bosses

Bosses who learn that an employee used AI may start second-guessing that worker’s claims. In a new set of experiments, supervisors accepted a worker’s word without double-checking 68 percent of the time when the worker had done a task alone. When AI was involved, that dropped to 24 percent. Supervisors in the main experiments never saw the work itself, only a software score saying whether AI had produced it, so quality could not explain the reaction.

AI chatbots have become a fixture of office life, and some workers use them in secret, yet little research has asked how bosses react once they find out. One thing eased the suspicion: a workplace that clearly welcomed AI. With that message in place, the gap shrank to something that could be chance. A ban did not help, and neither did saying nothing. Workplace expectations seem to matter, though the researchers did not test why bosses reacted this way.

Researchers at the Hebrew University of Jerusalem and WHU in Germany ran six studies, published in The Leadership Quarterly. Five were controlled experiments, most built around real-money choices that tracked what supervisors did rather than what they said, and one used real academic work.

Supervisors Trusted Workers Who Used AI Far Less

Study 1 included 392 adults who played temporary bosses. All held full-time leadership jobs in the United States, United Kingdom, or Canada, averaging about 43 years old with roughly a decade of supervising experience. Of them, 219 were men and 173 were women.

Each boss was paired with a real earlier participant who had written a social media post for a research-recruitment website, either alone, with AI, or by copying someone else. That worker had also reported how long the task took, and the report could be honest or padded. Bosses were never told the time. All they learned was what a detector said about the post, which stayed hidden.

Then came the money part. A boss could accept the report as is or pay 75 cents from a shared bonus pot to verify it. A padded report that slipped through cost the boss a bonus, so skipping the check meant gambling real money on the worker’s honesty. That choice stood in for trust.

Workers flagged for copying fared worst, earning trust only 11 percent of the time. The AI penalty held after the researchers accounted for bosses rating AI users as less competent and less professional, so skill alone did not explain it.

Task importance did not change the pattern. Bosses distrusted AI users whether the post was described as critical or as one of many outreach tools. Experience barely mattered either. Even people with no prior AI use who spent five minutes trying ChatGPT still distrusted AI users.

Real work told a similar story. In Study 6, 321 academic supervisors rated short introductions from real conference papers led by graduate students, each shown with an AI-detection score from the software Turnitin. The papers came from a 2024 management conference and were matched on topic and length, so those factors alone would not explain the results. The more of a paper the software flagged as AI-written, the less supervisors trusted its author. Supervisors with more AI experience were only slightly more forgiving.

Clear Policy Encouraging AI Use at Work Shrank the Trust Penalty

Study 2 told 601 supervisors that a worker’s company had banned AI, encouraged it, or said nothing. A ban and silence both left a heavy penalty, with trust in AI users at 19 percent under a ban and 33 percent under silence. When AI was encouraged, trust in AI users reached 47 percent, much closer to the 58 percent for workers who did the task alone. Workers under the ban were not blocked from using AI, and several likely did. One reasonable guess, untested here, is that bosses read AI use as cutting corners, and a clear welcome removes that worry.

Most companies still lacked a dedicated AI policy by late 2025, according to the paper, so many AI users may be working at companies that have never said where they stand.

Lower Trust Led to Lower Performance and Promotion Ratings

Study 5 gave 204 supervisors a real note from an earlier boss who either trusted a worker or double-checked that person’s claims, with no mention of AI. Readers of the trusting note rated the worker higher on job performance, being a good workplace citizen, and promotion potential. In Study 1, lower trust explained part of why AI users were rated lower.

These were experimental ratings, not real promotions, and bosses learned of AI use from software rather than from workers themselves. Whether admitting AI use would soften the penalty was not tested. The authors warn that workers who expect distrust may keep using AI but hide it.

Workers who use AI without a clear green light risk being doubted before their work is ever judged, and in these experiments, an employer’s silence did nothing to prevent it.

---

Disclaimer: This article summarizes one peer-reviewed study, in which participants played supervisor roles in controlled experiments, and is intended for general information only. Results may not apply to every workplace, and nothing here is professional, legal, or employment advice.

---