How does AI stigma differ from an AI skills gap?

Someone with a skills gap does not yet know how to use an AI tool, while someone held back by stigma already uses it and will not say so, because they expect a manager or client to judge them. In a session Anees Merchant ran in August 2026, a coach who called ChatGPT and Claude 'my employees' said they could not tell anyone how they used them. Anees Merchant's reading is that training does not move a status question.

Why do employees hide their AI use from managers?

Employees hide AI use because they expect to be judged for it. In four experiments with more than 4,400 participants, published in PNAS in May 2025 by Reif, Larrick and Soll, people who imagined using AI expected to be seen as lazier and less competent and said they would be less likely to disclose it. In the same experiments, observers rated hypothetical AI users as lazier, less competent, less diligent and less independent than people who got help from a colleague or no help.

When should a leader disclose their own AI use?

Anees Merchant's advice is to disclose early and specifically, before asking the team to do the same. This is his inference, and none of the experiments cited here tested it directly. In the Reif, Larrick and Soll experiments, managers who used AI themselves preferred candidates who used it, and frequent users were less likely to judge an AI user as lazy. Make the disclosure specific: name one piece of work, what the model did, how long it took, and which part stayed yours.

What is the first step to ending hidden AI use on a team?

The first step is for the most senior person in the room to describe one piece of their own work that AI helped produce, with the time it took and the part they did themselves. Follow it within the week with a short written list of tasks where AI use is expected. One experiment found the social penalty disappeared when a task was clearly suited to AI help, so stating the fit in advance is a reasonable step, though none of the studies cited here tested it.

Why Your Team Hides Its AI Use, and Three Moves That Make It Sayable

Q: What is AI stigma at work?

AI stigma is the fear of being judged lazy or less competent for using AI tools at work. It leads people to hide their use. In Slack's 2024 Workforce Index, 48% of 17,372 desk workers said they would be uncomfortable admitting to their manager that they had used AI for common tasks. In a 2025 University of Melbourne and KPMG study of more than 48,000 people, 57% of employees said they hide their AI use and present AI-generated work as their own.

AI stigma is the fear of being judged lazy or less capable for using AI at work. In a 2025 survey of more than 48,000 people, 57% of employees said they hide their AI use and present the output as their own. A coach put that fear to me in a live session in August and I answered it badly. Below are three moves a leader can make to get AI use said out loud. Two draw on published experiments, and one comes from that session.

The question a coach asked after my deck

One Sunday in August I ran a remote session on being limitless with AI for a room of entrepreneurs and solopreneurs, most of them coaches. One slide listed five self-limiting beliefs that keep people from using AI. The fifth was about delegation, and the line under it read: what will people think about me?

About twenty minutes later I opened the floor. The first question came from a coach who works alone. ChatGPT and Claude were "my employees", the coach said, and they did everything in the business. Then came this: "I feel that there is a stigma. I am not able to tell anyone about how I use it. Otherwise, I'll be judged."

Here was someone using AI at full depth who could not say so out loud.

My first answer was wrong, and I am keeping it in this essay for that reason. I reached for a technical fix. I described instruction files that tell a model to strip every sign of machine writing from its output, and I added that I often cannot tell AI content from human content myself. I was offering better hiding to someone who had just said they could not tell anyone.

The coach pushed back, and was right to. What bothered them was that the work would stop feeling like theirs, and that admitting it would cost the human connection a coaching practice runs on.

So I answered a second time. The deck I had just presented was built by AI. It took two hours, against the twenty it would have taken me by hand, and I said so to the room. Every story in that deck is mine. Then I gave the comparison I should have started with. If you hire a person to take your calls, nobody calls that shameful. The machine is doing a task for you, and the job is still yours.

The slide is the detail that matters most to me. I had put that exact fear on screen twenty minutes before the question, and the coach still asked it as though nobody else in the room carried it.

What the research says the coach got right

The coach's fear matches what people report in surveys and rate in experiments.

In May 2025 Jessica Reif, Richard Larrick and Jack Soll published four experiments with more than 4,400 participants in the Proceedings of the National Academy of Sciences. In the first, 497 people imagined finishing a work task with either a generative AI tool or a conventional one. Those who pictured using AI expected to be seen as lazier, more replaceable and less competent, and they said they would be less likely to tell anyone. In the second, 1,215 people rated descriptions of employees who got help from AI, from a colleague, or from nobody. The AI users were rated lazier, less competent, less diligent and less independent, and the result held whatever the employee's age, gender or job type.

The expectation in the first experiment matches the ratings in the second. People expect a penalty, and observers hand one out.

Two workplace surveys record what employees say they do about it. Slack's Workforce Index surveyed 17,372 desk workers in 15 countries in August 2024, and 48% said they would be uncomfortable admitting to their manager that they had used AI for common work tasks. The reasons they gave were a feeling that it was cheating (47%), fear of looking less competent (46%) and fear of looking lazy (46%). The University of Melbourne and KPMG surveyed more than 48,000 people in 47 countries between November 2024 and January 2025, and 57% of employees said they hide their AI use and present AI-generated work as their own.

In the same study 66% reported relying on AI output without evaluating its accuracy, and 56% reported making mistakes in their work because of AI. Training reached fewer than half: 47% of employees had received AI training, and 40% said their workplace had a policy or guidance on generative AI.

The KPMG release has one more number for anyone who signs off on risk. Almost half of employees admitted to using AI in ways that contravene company policies, including uploading sensitive company information into free public tools such as ChatGPT. KPMG reports the two figures separately and does not say how much they overlap. Where they do, a use nobody mentions is one nobody can check against the policy.

My own read of the coach's question is that training does not move a status question. The coach had no trouble with the tools.

A caution on all of these numbers. The first two experiments used imagined or described employees, and the surveys report what people say about themselves. None of them tested whether a leader's disclosure changes how a team behaves. That step is my inference from four experiments and one session.

Three moves that make AI use sayable

The Reif study also found conditions under which the penalty shrinks, and two of my three moves follow from them. The third comes from the session. All three belong to the person with the most standing in the room.

   MOVE 1             MOVE 2              MOVE 3
   Go first           Name what           Say where AI
                      stays yours         belongs
      |                   |                   |
  "Will I be          "Is this still      "Is this task
   judged?"            my work?"           allowed?"
      |                   |                   |
      +---------+---------+---------+---------+
                |
     AI use gets said out loud,
     so someone can review it

1. Go first, with numbers. In the third experiment, 1,718 people acting as hiring managers chose between candidates. Managers who did not use AI themselves tended to favour candidates who did not use it either, and managers with AI experience preferred the candidates who used it. The verdict tracked the judge's own AI habits. That is an experimental result, so what follows is my inference: when the leader is also the judge, a leader who uses AI and says so lowers the penalty the team expects. My second answer in the session was a specific disclosure about one piece of work: this deck, two hours against twenty, built by a model.

2. Name what stays yours. This move has no experiment behind it. It comes from the coach, whose fear of judgment sat on top of a question about authorship: would the work still feel like theirs? The answer I gave was the hired assistant taking calls. Delegating a task is ordinary, and it leaves the judgment and the accountability with the person whose name is on the work. In my deck the model built the slides, and every story on them came from my own life. Your team needs the same sentence for their own work, in their own words, because a client or a colleague may ask, and they will want an answer they can defend.

3. Say where AI belongs. In the fourth experiment, 1,006 participants evaluated candidates for manual and digital tasks. When the task was digital and clearly suited to AI help, the penalty disappeared. The study also reported that frequent AI users were less likely to judge AI users as lazy. In the KPMG data only 40% of employees said their workplace had a policy or guidance on generative AI. My inference is that a short written list of tasks where AI use is expected gives the rest something to check a use against, and no study I have cited tested that. The list has a second use. A manager cannot review work she does not know was drafted by a model, and the list tells her where to look.

The answer I would take back

My first answer in that session deserves a second look, because it is the tempting one. When someone says they are afraid of being judged, a tool that hides the evidence feels like help. It tells them the fear is correct and the use is something to conceal.

In my view a leader gives the same answer by saying nothing, when no one asks how a piece of work was made and no one volunteers it. Slack's survey found a link between comfort and use. Workers comfortable sharing that they had used AI were 67% more likely to have used it for work than those who were not. The survey cannot say which one causes the other.

Three things to do this week

Disclose one piece of your own work. At your next team meeting, name one thing you produced with AI in the past month. Say what the model did, how long it took, and which part was yours. Keep it to one example and give the hours, as I did with the deck: two against twenty.
Publish a five-task list. Write down five tasks on your team where AI use is expected, and send the list by Friday. Start with digital tasks clearly suited to AI, since that is where the fourth experiment found the penalty disappeared. Add one line on what the person signing the work remains accountable for.
Ask the question in writing. Send one anonymous question: what do you use AI for that nobody here knows about? Use a form that does not record names. Count the answers and read every one. The count gives you a floor on reported use, since anyone who does not answer stays invisible.

Footer

Anees Merchant writes one essay every Tuesday about enterprise AI, agentic systems, and the human side of the work. He is the author of Merchants of AI, a TEDx speaker, and a doctoral researcher in Human-AI Communication at the Swiss School of Business and Management.

The newsletter version, with extra commentary, goes out separately on Mondays. Subscribe at /newsletter.

See you Tuesday.