Workers scrub AI fingerprints from their work before showing colleagues — but does that trick actually fool anyone?

This explores whether hiding the traces of AI use actually earns workers a competence boost from colleagues, or whether that boost is mostly assumed. The direct answer is that the corpus doesn't measure peer reactions to concealed AI use. What it documents well is why workers conceal it, and what that concealment costs. Interviews with knowledge workers show they strip out GenAI cues for two reasons: to avoid stigma, and because a clean deliverable makes them look like the expert Why do knowledge workers hide signs of using GenAI?. Large experiments back up the fear behind this. People using AI expect to be rated as less competent and less diligent, and they become less willing to tell managers or colleagues about it Do people fear judgment when they use AI at work?. Note that this research measures the penalty people *expect*. Whether peers actually reward the hidden version is still an open question here.

One reason concealment might work is that polish is persuasive. A study of models trained to imitate ChatGPT found they fooled human evaluators by copying its confident, fluent style without getting any more accurate Can imitating ChatGPT fool evaluators into thinking models improved?. Peers judging a smooth deliverable face the same problem. They see the finished surface, not the work behind it. Another study of how AI reshapes workplace signals found that workers carefully keep cues tied to identity, like their own voice and who authored what. They let cues about effort, attention and uncertainty quietly disappear into the output Which workplace cues survive AI mediation and which disappear?. Removing the AI traces therefore also removes the information colleagues would normally use to judge how hard something was, or how sure the author is.

The less obvious twist is that the same illusion can work on the person doing the hiding. Research on fluency shows that polished AI output makes users feel more capable themselves, even though they didn't produce it Does processing ease mislead users about their own competence?. Researchers call this the "LLM Fallacy": people credit their own ability with what the AI contributed. It is a separate problem from hallucination or over-trusting the tool How does AI-assisted work reshape how people see their own abilities?. A worker who erases GenAI cues may therefore be managing two impressions at once, their colleagues' and their own. Neither has to match their actual skill.

Concealment also has a cost at the team level. When everyone hides their AI use, colleagues stop learning from each other's workflows, and the culture of not being open about GenAI reinforces itself Why do knowledge workers hide signs of using GenAI?. Meanwhile, what counts as competence is itself shifting. Workers describe four different outcomes: skills that develop, skills that are maintained, skills that erode, and skills whose value changes How does generative AI actually change worker skills?. Heavy users of Claude report feeling more optimistic about their careers, though that evidence is only a correlation within Anthropic's own user base Does delegating work to AI actually damage worker skills?. So the short answer is: possibly, at least in the short run, because fluent output is persuasive. But the competence being signalled may not be the competence that exists, and the corpus has yet to test whether peers are actually convinced.