Thousands of students in Hamilton County Schools participated in two studies looking at how artificial intelligence tools affect learning, with researchers saying student success hinged less on access to AI tools than how they ultimately were used.
The first of the studies, both of which were lead-authored by University of Toronto researcher Philip Oreopoulos, gave students in 18 of the district’s middle schools access to a Khanmigo AI math tutor produced by Khan Academy. The study featured 2,708 students and ran across the 2024–25 and 2025–26 school years.
Within each of the schools, one or two grade levels were given access to the Khan Academy learning suite and the Khanmigo tutor during daily remedial math practice, while other classes relied solely on teacher-led practice and alternative software not powered by AI.
What researchers found was that while the AI-supported students’ scores improved slightly over their peers who didn’t have access to the Khan Academy tools, the gains had little to do with the AI tutor itself. Of all the students surveyed during the second year, 96% messaged Khanmigo at least once, but the median student during the whole of the study didn’t use the tutor on two-thirds of all days and only used it in 17% of exercise sessions during which they made a mistake.
The solution, Oreopoulos said, isn’t just giving students access to AI, especially when its value isn’t immediately obvious to them.