Should I tell my boss I used AI?

Yes, when the AI touched work you will be judged on, and say what it did as well as that you used it. The evidence is mixed in a useful way. In four experiments published in PNAS in 2025, people described as using AI for a task were rated lazier and less competent than people described as getting the same help from a colleague or another tool, so the fear of disclosing is not imaginary. The same experiments found the penalty was largest among evaluators who rarely used AI themselves, and that it disappeared when the tool's usefulness for the specific task was made explicit. Surveys from Microsoft in 2024 and Deloitte UK in 2026 show large minorities of workers hiding their use anyway. The ratings in the experiments were about described strangers, and nobody has measured what happens to real employees who disclose or conceal.

What four experiments found about being judged#

Reif, Larrick and Soll ran four preregistered online experiments with over 4,400 participants, per the abstract. Participants who imagined using generative AI expected to be seen as lazier and more replaceable, and said they would be less willing to tell managers and colleagues. Evaluators then rated described employees who received AI help against those who received other help or none, and the AI group was rated lazier, with lower marks on competence, diligence, independence and self-assurance. There was no difference on ambition or dominance. Perceived laziness carried the effect.

Two findings cut against the gloomy reading. In the hiring study, managers who rarely used AI themselves preferred candidates who reported never using it, while managers who used AI frequently preferred candidates who used it daily. And when the task was digital and the usefulness of AI for it was made explicit, the penalty disappeared: candidates using traditional tools and those using AI tools were rated alike on fit for the task.

Two large surveys show many workers keeping their use quiet#

Microsoft and LinkedIn's 2024 Work Trend Index, a survey of 31,000 knowledge workers in 31 countries fielded in February and March 2024, reported that 52 per cent of people who use AI at work were reluctant to admit using it for their most important tasks, and 53 per cent worried that using it on important tasks made them look replaceable. Deloitte UK's 2026 survey of 25,000 British workers found that among users, 31 per cent use generative AI without their employer's knowledge, 23 per cent perceive a stigma around it, and 64 per cent of weekly users worry their manager will think AI can do their job. Both are self-report, and both come from firms that sell or advise on the adoption they are measuring.

What the experiments cannot say about your workplace#

• Described strangers, not colleagues. The experiments used vignettes and candidate profiles. A manager who knows your work may judge differently, and no study here observes real reactions.
• Nothing measures the cost of concealing. The evidence prices the penalty for being seen. It has no figure for what follows when a manager, a client or a log later shows that AI was used and was not mentioned.
• The penalty may be shrinking. The authors expect norms to shift as use becomes ordinary, and the surveys are one to two years apart.
• Policies differ. Some employers require disclosure, some restrict tools, and the studies say nothing about which applies to you.
• Silence is not always visible only to you. On an employer-managed account, administrators may be able to see the prompts, as the vendors' own documentation describes.

Say what the tool did and what you did#

This section is interpretation, kept apart from the evidence above.

The experiments that found a penalty described AI use as a bare fact. The condition that removed it added a reason the tool fitted the task. That points at the sentence to write, which has two halves: what the tool produced, and what you decided, checked or changed. It is the same test how to be honest about using AI sets for students, whether you could describe what you did without leaving bits out.

Rahim Hirji made the reader's side of this argument in Leave the Fingerprints In on 14 June 2026. His observation is that readers have learned to sense synthetic prose, and that machine polish "no longer looks like a shortcut" but like care, which hides that nobody did the work. Put beside the Duke finding, the risk of a bare disclosure is that it invites the laziness inference, and the risk of silence is that the polish gets read as effort you did not spend. Naming your own contribution puts a person back into the work. That reading is his, and the experiments did not test it.

A practical way to tell them#

• Check the rules first. If your employer has a policy on AI use, follow its disclosure terms, and if it has none, ask what is expected.
• Disclose the use that matters. A spell-check does not need announcing. A draft, an analysis or a recommendation that shaped a decision does.
• State fit and contribution together. "I used the model to produce a first pass of the comparison table because it is a formatting-heavy task, then checked every figure against the source and rewrote the conclusion." Both halves are in that sentence.
• Prefer a channel you control. Say it in your own message or covering note, before someone finds the pattern in a log or in the prose.
• If you manage people, say out loud that disclosed use is not judged as laziness, and state what you do want disclosed. In the Microsoft survey about half of AI users said using it on important work made them look replaceable, and a manager can answer that worry directly.

Key sources

• Reif, J. A., Larrick, R. P. and Soll, J. B. (2025). Evidence of a social evaluation penalty for using AI. PNAS 122(19), e2426766122. Four preregistered experiments, over 4,400 participants. Graded entry.
• Microsoft and LinkedIn (2024). 2024 Work Trend Index Annual Report. 8 May 2024. 31,000 knowledge workers, 31 countries. Graded entry.
• Deloitte UK (2026). British workers spend one billion pounds of their own money on gen AI for work. 16 September 2026. 25,000 UK workers. Graded entry.
• Hirji, R. (2026). Leave the Fingerprints In. Box of Amazing, 14 June 2026. Argument, cited as interpretation rather than evidence.

Related SuperSkills research#

On what an employer can see, can my employer see what I put into AI. On honest description of your own use, how to be honest about using AI in your work. On the fear underneath, does using AI make you less worried about your job. On what leaders should say, what to tell employees about AI and headcount. On personal messages, should I use AI to write personal messages.

About this research#

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. The PNAS paper's authors, sample sizes and findings were read at source on 29 September 2026, and the Microsoft figures on the same day; all three studies are graded in the evidence base. The social evaluation penalty is Reif, Larrick and Soll's finding and not a SuperSkills term.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.