When coworkers openly use AI, do their colleagues really think less of them, or is that penalty mostly what AI users expect?

This explores whether people who see a coworker using AI actually think less of them, or whether the penalty mostly exists in AI users' expectations. The short answer from this collection is that the expectation is well documented, but the observers' side is thinner than the question assumes. The strongest evidence comes from four experiments with more than 4,400 people. AI users expected to be seen as less competent and less diligent, and they became less willing to tell managers and colleagues they had used the tool Do people fear judgment when they use AI at work?. That is a measured fear, and people act on it. As summarized here, though, it does not show how harshly observers really judge.

The collection is also frank about what nobody knows yet. One note asks whether the penalty will fade once AI is as ordinary as spellcheck, and finds no data tracking it over time Does the social penalty for AI use fade as the tool becomes ordinary?. It also raises a less obvious possibility. The judgment may not depend on novelty. It may depend on agency: AI does part of the work, so observers may keep asking how much of the result is really yours, however familiar the tool becomes. If that's right, the penalty is about who gets credit, and time alone won't remove it.

The more surprising material is about what workers do with that anticipated judgment. An analysis of 1,250 worker interviews found that people carefully keep the cues that mark work as theirs, such as their own voice and where ideas came from. Meanwhile, signs of effort, attention and uncertainty quietly disappear into polished AI-assisted deliverables Which workplace cues survive AI mediation and which disappear?. So the social penalty may be changing what observers can see at all. If finished output counts as proof the work happened, a colleague has less to penalize, and also less to go on.

That hiding has a cost for the user too. Another line of research describes an "LLM fallacy": when AI-assisted output is smooth enough, people count it as evidence of their own skill and believe they can do things they can't Do AI-assisted outputs fool users about their own skills?. Put that next to the disclosure findings and a loop appears. Fear of judgment pushes AI use out of sight, and when the line between human and AI work is hidden, it gets blurrier for observers and users alike. A related study of dishonesty shows how strongly human judgment shapes behavior: people inclined to cheat prefer reporting to a machine rather than a person, because a machine doesn't judge Do dishonest people prefer talking to machines?.

The nearest thing here to observers actively penalizing AI use is hiring. In a Greenhouse survey, 91% of recruiters say they spot deception in applications, and 41% of job seekers say they use AI prompt injections to slip past screening filters Are job applicants and employers locked in an escalating AI arms race?. That is a penalty for gaming the system, not for openly using AI, so it doesn't settle the question. To find out whether observers really mark people down, and by how much, you'd need studies that measure the evaluators directly. Based on these summaries, this collection doesn't have one yet.