More than half of students using AI may not report it, new study finds

A UC Irvine School of Education-led study finds evidence that some students may be relying on generative AI in ways that short-circuit learning, and that traditional survey methods may underestimate the extent of the trend.

By Rachel Sampson
April 13, 2026

As generative AI becomes a routine part of academic life, a new UC Irvine-led study is asking a deeper question: Are students using AI to support learning, or avoid it?

In a working paper, researchers from the UC Irvine School of Education and collaborators examined undergraduate students’ use of AI through actual coursework behavior, including written responses, assignment logs, and behavioral indicators, rather than relying only on self-reported surveys.

What they found points to a more complicated picture.

​The study found that 41% to 70% of students showed at least one sign of what researchers call “Concerning AI Usage,” depending on how conservatively the behavior was measured. The study also identified a notable gap between what some students reported about their AI use and what the behavioral data suggested. In fact, more than half (52%) of students who showed behavioral signs of concerning AI use reported that they had not used AI at all, underscoring the limits of self-reported data. 

Sina Rismanchian

“Our project originally began from optimism about what large language models can offer for education,” said Sina Rismanchian, Ph.D. student and lead researcher on the project. “But before building solutions, we felt it was essential to understand the current reality: how students are already using these tools in their everyday learning.”

​That gap between perception and reality is part of what makes the study stand out.​

​“While survey studies suggest students are using AI tools in large numbers, they also tend to report that students rarely have AI complete an entire assignment—and generally believe that doing so is wrong,” said School of Education assistant professor Shayan Doroudi. “What's different about our study is that we directly measure students' AI usage in context.”

​Doroudi added that traditional approaches may miss key behaviors.

Shayan Doroudi

“In my experience as an instructor, students may use AI in problematic ways but are unlikely to admit to it,” he said. “Our findings highlight a clear discrepancy between self-reported behavior and what students are actually doing.”

The findings also raise questions about how students perceive their own learning and academic behavior. Students who showed the highest levels of AI use self-reported higher levels of intellectual integrity on average than those who showed no signs of AI use - scoring 3.97 out of 5, compared to 3.38 among students with no AI flags - suggesting a potential disconnect between students’ self-perception and their actual practices.
​
For the research team, the issue extends beyond academic integrity. It raises broader questions about learning itself and what may be lost when students rely on AI to bypass the effort required to engage with course material.

“AI tools are incredibly convenient, especially when coursework feels difficult or overwhelming,” Rismanchian said. “But struggle is often where real learning happens. Avoiding that struggle can come at a long-term cost.”

Rather than focusing solely on restriction or enforcement, the team is now looking ahead. Their next phase of work centers on collaborating with students to design learning experiences that encourage more meaningful engagement with AI—and cultivate intellectual virtues such as curiosity, humility, and intellectual tenacity.

The working paper is available as a preprint on EdArXiv. Read the full study here.

---

This research was supported by Grant 63365 from the John Templeton Foundation. The opinions expressed are those of the authors and do not necessarily reflect the views of the Foundation.