Would your team tell you they used AI? The hidden trust penalty at work

Same work, different verdict

Picture two people on the same team handing in the same report. One wrote it alone. The other used an AI assistant. The finished work is equally good.

Which of them does the manager trust more?

A new peer-reviewed study suggests the answer is uncomfortable, and that it matters for every organisation trying to get real value from AI.

What the research found

The study, published in The Leadership Quarterly, was led by Samuel Meyers and Dr Yochanan Bigman of the Hebrew University of Jerusalem with Dr Roman Briker of WHU, the Otto Beisheim School of Management in Germany. Across six studies, the researchers looked at how supervisors react when they know an employee used a large language model such as ChatGPT.

In one experiment, participants acting as supervisors had to decide whether to accept an employee's report of how long a task took, or check it themselves. When the employee had done the work independently, they had a 68% chance of being trusted. When the employee had used AI, that fell to 24%. The supervisors were not judging the quality of the work, only how it had been produced.

Further experiments linked that lower trust to weaker evaluations of performance, a lower view of the employee's contribution, and lower perceived suitability for promotion. A separate study of academic supervisors assessing real work by graduate students found the same direction: the more students used AI, the less their supervisors tended to trust them.

Two findings stand out for leaders. First, the trust penalty appeared when AI use was prohibited and when there was no policy at all. Where the organisation explicitly encouraged AI use, the effect was substantially reduced and no longer statistically significant. Second, neither the importance of the task nor the supervisors' own experience with AI consistently removed it. Using AI yourself does not, on this evidence, stop you marking others down for it.

The researchers warn that this could push people to conceal their AI use, often called shadow AI.

The other half of the picture

A survey by JobAssist, reported by TechRadar on 9 November, shows what that pressure can look like from the other side. It found that 52% of workers have used AI on a task they believed they could finish faster without it, and 63% say their employer encourages or expects AI use for at least some tasks. 39% have highlighted their AI use to managers even when they did not think it helped, 31% kept using it after deciding it had not improved speed or quality, and 34% held back negative feedback about AI. The coverage does not give the sample size, country or dates, so treat these figures as indicative.

Put the two together and a pattern appears. Some people hide their AI use because they fear being judged for it. Others perform it because they fear being judged for not using it. Both are responding to the same thing: they cannot tell what their manager really thinks, so they guess.

When people guess, leaders lose sight of the truth. The work arrives without anyone saying how it was made, and the reports about AI arrive polished, with the disappointments quietly removed.

What it means for leaders

Ian Hughes makes the case for openness in his book What Stays the Same, writing about clients: "A client will usually forgive you for using a machine. They will struggle to forgive you for lying about it." He also notes: "Openness looks like a cost. It behaves like an advantage."

The same logic works inside a team, and in both directions. Staff need to feel safe saying where AI helped. Leaders need to hear, just as safely, where it did not.

Practical steps

1. Say it out loud. The study suggests silence works like a ban. Tell people where AI use is welcome, where it is not, and why.
2. Judge the work, then ask about the method. Check the output against the standard first. Treat AI use as information about how the work was done, not a mark against the person.
3. Agree how AI-assisted work is checked. A simple, shared expectation for review gives managers a reason to trust the result rather than relying on instinct.
4. Make "it didn't help" a useful answer. Ask teams regularly where AI saved time and where it cost time, and thank people for the second as much as the first.
5. Measure outcomes, not usage. Counting prompts rewards the show. Ask what changed for the customer or the service.

Where training fits

None of this is technical. It is about how managers judge people, how they set expectations, and whether their teams feel able to tell them the truth. Those are leadership skills, and they can be learned. Line managers in particular benefit from practising how to talk about AI use with their teams, how to review AI-assisted work fairly, and how to ask for honest feedback without it sounding like a test. Service management thinking helps too, with its long habit of measuring outcomes rather than activity and of building feedback into the way work is run.

Honest limits

Most of the evidence in the Leadership Quarterly paper comes from experiments in which participants played the role of supervisors, alongside one study of academic supervisors. Real workplaces are messier, and the size of the effect may differ. The JobAssist figures come without published methodology in the coverage we saw. And any policy on AI use at work should be checked with HR and, where needed, legal advice, especially where it touches performance reviews or promotion.

The direction is still clear enough to act on. If people cannot tell how you will react, they will either hide their AI use or stage it. Neither tells you what you need to know.

Tags: ai adoption, leadership, trust, shadow ai, line managers, psychological safety

Sources

• https://www.afhu.org/2026/11/09/using-chatgpt-at-work-your-boss-may-trust-you-less-new-study-finds/

• https://doi.org/10.1016/j.leaqua.2026.102004

• https://www.techradar.com/pro/workers-feel-pressure-to-use-ai-and-show-off-to-bosses-even-when-it-slows-them-down