AI Teaching Assistants Lower Student Motivation, Study Finds

A randomized controlled trial involving nearly 3,000 students has found that giving teachers access to AI assistants can backfire, reducing student motivation and in some cases lowering test scores.

The study, conducted by researchers at the University of Pennsylvania's Wharton School, tracked 193 teachers and more than 2,800 middle and high school students in Turkey over 10 weeks during spring 2025. Teachers randomly assigned to use a ChatGPT-based assistant customized to Turkey's national curriculum primarily used it to generate lecture notes, assignments and exams.

Students whose teachers had AI access rated their classes as less enjoyable, less interesting and less important compared to control groups. While overall academic achievement remained flat, students of lower-performing teachers—identified by their students' prior test scores—saw both achievement and confidence decline on externally administered standardized exams.

Why it matters

This represents one of the first rigorous experimental tests of AI teaching tools in real classrooms, and the results challenge the widespread assumption that AI will automatically make teachers more effective. As school districts invest in AI platforms, the findings suggest that simply providing access to generative AI without training or guardrails may harm rather than help instruction, particularly for teachers who need the most support.

The delegation problem

"Teachers, just like students or coders, might be using AI as a crutch," said lead author Alp Sungu, an assistant professor at Wharton. "Instead of doing the actual work, they're using AI to delegate the task, and that lowers the quality of their teaching."

The research team did not directly observe classrooms or analyze the AI-generated materials teachers used, so they cannot pinpoint exactly why teaching quality deteriorated. Sungu suspects teachers may be losing their personal voice and style by relying on uniform AI-generated content.

Stronger teachers, he theorizes, likely treat AI output as a first draft requiring substantial revision and adaptation. Weaker teachers may use AI-generated materials with minimal modification, essentially outsourcing core instructional design work.

Not a time saver

Sungu himself uses AI in his university teaching to create interactive elements that would otherwise be too time-consuming. But he finds the initial output often contains errors or poorly chosen examples that require extensive correction. "I end up spending an equal amount of time to improve the output or calibrate it to my class," he said. "It's not a time saver."

The study design actually understates potential risks, since control-group teachers were free to use other AI tools on their own. The comparison measured the effect of providing a customized AI assistant versus whatever teachers chose independently.

Sungu emphasizes the findings should not lead to blanket conclusions that "AI is terrible and will ruin education." Instead, the lesson is that access to AI technology alone does not improve teaching. Effective use requires training programs, guardrails and better interfaces that preserve human judgment and creativity.

The draft study, titled "Generative AI Can Harm Teaching," was released online in June and has not yet been peer-reviewed. Details were first reported by The Hechinger Report.