Artificial intelligence tutors are increasingly finding a place in university classrooms, offering students instant explanations, academic assistance and help with coursework. However, a new study from researchers at the University of Maryland has raised questions about whether these tools are improving academic performance or reducing students’ engagement with traditional learning activities.The study, released on September 30, examined the impact of a course-integrated AI tutor on 2,379 undergraduate students across 30 instructors. Researchers found that students who were given access to the AI assistant recorded lower final grades and participated less in instructor-designed online learning activities than students who did not receive access.According to the research findings, students in sections provided with the AI tutor recorded final grades that were lower by 0.27 to 0.37 standard deviations. When researchers compared sections of the same course, the estimated decline was 0.37 standard deviations.Student participation in online learning activities also fell by 0.90 standard deviations, raising questions about how AI-based academic assistance influences students’ involvement in coursework.The research, titled “The Effects of Course-Integrated AI Tutoring on Student Performance and Engagement: A Randomized University Trial”, was published as an EdWorkingPaper by the Annenberg Institute at Brown University.
What did the University of Maryland study find?
The researchers conducted a full-semester randomised controlled trial involving undergraduate courses across science, technology, engineering and mathematics (STEM), social sciences, humanities and education.Students in selected course sections received access to an AI-powered study assistant integrated into the university’s learning management system. The tool was based on GPT-4o and used retrieval-augmented generation to provide responses grounded in course materials selected by instructors.The researchers compared their academic performance and engagement with students who did not receive access to the assistant.According to the findings, students with access to the AI tool visited course pages less frequently, recorded fewer active days and completed fewer discussion responses, assignments and quizzes.Homework scores also showed no significant improvement or negative effects, rather than the short-term academic gains reported in some earlier studies.Survey responses further indicated that students using the assistant asked instructors fewer questions about course content.The researchers noted that introducing an AI tutor could displace engagement with existing learning activities without necessarily improving academic performance.
First-generation students recorded larger grade declines
One of the findings highlighted in the research concerned differences in academic outcomes among student groups.According to the study, the estimated decline in final grades among first-generation college students was more than twice that recorded among students whose parents or families had previous experience in higher education.The researchers raised concerns about the possibility of unequal effects across student groups, particularly when AI tools are introduced without sufficient academic guidance.The findings are relevant to universities considering AI-based learning support as a way to improve access to academic assistance.
How students used the AI tutor
The study also examined the types of requests students made to the AI assistant.According to the researchers, approximately 73.8 per cent of requests sought information, explanations or solutions. Around 11.3 per cent involved practice or test preparation, while only 0.7 per cent requested feedback on students’ own work.The assistant offered instructors two operating modes: direct instruction, which was the default setting, and a tutor mode designed to guide students towards answers rather than provide solutions immediately.However, the researchers reported that very few instructors changed the default setting. Only two instructors had switched to tutor mode by the end of the semester.This distinction is important to understanding the study. The researchers examined an AI assistant as it was actually deployed, rather than a system consistently configured to encourage students to work through problems independently.
Researchers caution against generalising the findings
The study also identified limitations that need to be considered when interpreting its results.Only around 15 per cent of students in the treatment sections used the university’s AI assistant at least once. The researchers noted that the introduction of the tool may also have influenced how students used other publicly available AI platforms, including ChatGPT and Gemini, which remained accessible to all students.The authors also clarified that their conclusions apply to the particular AI assistant and configuration examined in the study. They did not establish that all AI tutoring systems negatively affect academic performance.Earlier research has also explored the importance of how AI learning tools are designed. A 2025 study involving high school mathematics students, for instance, found that unrestricted access to general-purpose AI could affect subsequent independent performance, while additional tutoring safeguards helped mitigate those effects.The Maryland findings therefore add to ongoing research into the relationship between AI-assisted learning, student participation and academic outcomes.
What the findings mean for universities
The research comes as universities increasingly explore AI-powered tools for classroom assistance, personalised learning and academic support.While AI tutors can provide students with access to explanations and information outside classroom hours, the Maryland study raises questions about their implementation, particularly when the tools provide direct answers instead of encouraging independent problem-solving.The researchers’ findings are based on one university’s full-semester trial and do not establish that AI tutors universally reduce grades. However, they provide evidence of how the introduction and configuration of an AI assistant can influence student performance and participation.The complete research paper, “The Effects of Course-Integrated AI Tutoring on Student Performance and Engagement: A Randomized University Trial”, is available through the Annenberg Institute’s EdWorkingPapers platform.