Students overestimate AI use, fueling silent pressure

students overestimate – A University of Chicago-led study suggests students admit to using AI far less than they believe their peers do—an attitude shift that can quietly turn into pressure to “keep up,” even when many classmates aren’t using large language models to complete coursew

On a quiet campus moment, the math can feel personal: students aren’t just deciding whether to use AI tools. They’re also deciding what their peers will think if they admit it.

“Students don’t want to be perceived by their peers as not able to do the work. ” said Alex Kale. a computer scientist at the University of Chicago and a co-author of the study. which was presented at a conference in Barcelona. Spain. in April. “They don’t want to be perceived by their peers as dishonest … And it feels deeply personal.”.

Kale described the dynamic as “social desirability bias. ” the human tendency to answer questions in a way that makes us look good to others—rather than being fully candid—even in an anonymous survey. In a separate online survey of 98 undergraduates conducted by the researchers. respondents said admitting to using AI was similar to admitting they were “not able to complete coursework independently. ” or were “lazy.” One respondent also believed students were hiding their usage because they feared getting caught and possibly expelled.

But the researchers say the gap may not only come from what students choose to admit. They may also be misreading how common AI use really is.

Students, the study suggests, could be overestimating how many of their peers use AI because the tools are visible across campus life. They hear people talking about ChatGPT. They see AI tools open on laptop screens. That constant exposure can make it feel like the norm—even if it isn’t.

One survey respondent put it this way: “I think only a small portion of students actually rely on LLMs to do coursework. while most students do not. That small portion leads some students to assume most are using it.” The post-2022 generation of AI tools like ChatGPT are often referred to as large language models. or LLMs.

Taken together, the study’s story is uncomfortable in its simplicity: students may be using AI more than they admit, while the hype around AI may also be creating the impression that everyone is using it.

The pattern is familiar in other areas of student behavior research, particularly public health studies about alcohol, drugs, and sex. There, researchers have long documented a gap between what people report doing and what they believe their peers are doing. Students often overestimate how much others drink heavily, use drugs, or have casual sex. And those false perceptions have had real consequences for attempts to reduce unhealthy behavior.

When students believe that “everyone else is doing it,” the belief can start to work like a push. It becomes partly self-fulfilling.

More than 25 years ago. colleges began to worry that warning students about binge drinking on campus was backfiring—encouraging them to get drunk instead. Many shifted strategy. They downplayed binge drinking and publicized statistics showing that most students drink in moderation. In reporting from some public health officials, the number of students who said they drink heavily declined.

There may be lessons in that experience for how educators encourage responsible AI use, even though the University of Chicago study doesn’t link AI use to drugs or booze. The study’s central message is narrower, but it lands harder: perceptions matter.

If students believe that nearly everyone is relying on AI to complete coursework. the pressure doesn’t have to come from teachers or administrators. It can come from the student’s own sense of what “normal” looks like. And that belief—however inaccurate—can nudge decisions in the direction of conformity.

Kristin Fasiang is a graduate student in computer science and learning sciences at Northwestern University. Fasiang reported and wrote this story along with The Hechinger Report’s Jill Barshay.

AI use social desirability bias ChatGPT large language models LLMs college students academic integrity peer perceptions Northwestern University University of Chicago Hechinger Report