Most studies on AI in education are snapshots. A one-week experiment, a post-survey, a quick reaction from students after their first interaction with ChatGPT. Useful, sure. But they miss the longer arc of how students and instructors actually adjust to AI once the novelty wears off and the real work begins.
Guo, Li, and Cunningham (2025) did something different. They followed two courses, one in psychology and one in physics, across an entire academic year. Students used ChatGPT as a learning tool throughout the year, and the researchers tracked how perceptions shifted over time. The result is one of the more grounded accounts I’ve seen of what happens when ChatGPT moves from a one-off experiment to a regular presence in the classroom.
ChatGPT in the classroom: Students Liked It, But They Didn’t Want It Replacing Their Professors
The overall picture was positive. Students found ChatGPT easy to use, enjoyed interacting with it, and saw it as a helpful supplement. As the authors report: “Overall, students found ChatGPT easy to use and beneficial as a resource for learning. They enjoyed interacting with the tool, felt it boosted their confidence and motivation, and saw it as a valuable classroom supplement” (Guo et al., 2025, p. 7).
But here’s the part that should reassure every educator reading this: students consistently rejected the idea that ChatGPT could meet their individual needs better than their instructors. “At the same time, the students did not believe that ChatGPT could replace the instructors in classrooms to address their individual needs” (Guo et al., 2025, p. 7).
That’s a finding worth sitting with. Students appreciated what AI added to their learning experience, and they were clear-eyed about what it couldn’t do. The human instructor still held a role that ChatGPT couldn’t fill, particularly when it came to personalized guidance, nuanced feedback, and understanding individual learning situations.
The Activity Type Made All the Difference
One of the strongest findings in the study was that the type of learning activity shaped student perceptions dramatically. In the psychology course, students used ChatGPT for structured, skill-based coding tasks. In the physics course, the activities were more conceptual, centered on discussion and abstract reasoning.
The coding tasks won. Students responded far more positively when they could use ChatGPT to test, revise, and iterate on concrete problems. Guo et al. explain: “These findings suggest that the type of activity plays a crucial role in shaping students’ perceptions. Coding activities may have been perceived as more valuable because they offer more tangible and structured learning opportunities compared to the abstract nature of conceptual discussions” (2025, p. 10).
This makes intuitive sense. When a student writes a piece of code and can immediately run it, see whether it works, and then ask ChatGPT to help debug or improve it, the feedback loop is tight and tangible. The student can see the result of the AI’s help in real time. Conceptual discussions don’t offer that same concreteness.
A ChatGPT explanation of a physics principle might sound complete, but a student has no easy way to test whether they actually understood it or just read something that sounded convincing. That’s where the risk of cognitive surrender comes in, the tendency to accept AI-generated explanations without genuinely processing them, which Shaw and Nave (2026) documented in their research on how AI reshapes human reasoning.
Perceptions Changed Over the Year
The longitudinal design of this study revealed something that short-term studies can’t: student perceptions of ChatGPT improved over time, particularly in the physics course. Early in the year, students in physics were less enthusiastic about using ChatGPT for conceptual tasks. But once the instructors shifted the design to emphasize group work and self-exploration, engagement picked up.
This is a critical point. The tool didn’t change. The pedagogy did. And when the pedagogy improved, so did the student experience. Guo et al. are direct about this: “These findings suggest that while ChatGPT may be a valuable educational tool, its effectiveness depends on the context in which it is used, the nature of the learning activity, and students’ perceptions of its role in their education” (2025, p. 12).
I’ve been making this argument across multiple posts now. AI’s value in education is almost entirely a function of how it’s integrated. The tool on its own does nothing. The pedagogy and the teacher’s guidance determine the outcome.
Graduate Students Got More Out of It
An interesting secondary finding: graduate students showed higher motivation and stronger intentions to keep using ChatGPT compared to undergraduates, especially for skill development. They saw the tool as a professional asset they could carry beyond the classroom.
What’s also worth noting is that demographic variables like gender, race, and first-generation status showed no statistically significant differences in how students perceived ChatGPT. When access and support are in place, the benefits appear to be broadly distributed. That’s an equity finding that deserves more attention in future research.
The Instructor Perspective
The study also captured instructor reflections, and they paint a balanced picture. Both instructors saw real value in ChatGPT for supporting engagement and exploration. But they also flagged clear limitations. Without careful scaffolding, prompt guidance, and technical support, students risked shallow understanding, misinterpretation, or overconfidence in what ChatGPT told them.
That tracks with everything the research has been showing us. AI can be a powerful classroom tool, but only when an instructor has thought carefully about how to embed it into the learning design. Drop ChatGPT into a course without that structure and you get surface-level engagement at best, and misplaced confidence at worst.
Why This Study Matters
A full academic year of data across two disciplines with both quantitative and qualitative analysis. That’s rare in the current AI-in-education literature, and it gives this study a weight that many one-shot experiments lack. The findings reinforce a pattern that keeps showing up across the research: AI’s impact on learning depends almost entirely on the instructional design around it. The tool is only as good as the pedagogy that shapes its use.
For anyone still wondering whether ChatGPT has a place in the classroom, the answer from this study is clear. Yes, it does. But that place has to be designed with intention, guided by an instructor, and matched to activities where students can engage with AI actively and critically.
Reference
Guo, F., Li, T., & Cunningham, C. J. L. (2025). One year in the classroom with ChatGPT: Empirical insights and transformative impacts. Frontiers in Education, 10, 1574477. https://www.frontiersin.org/journals/education/articles/10.3389/feduc.2025.1574477/full