Can structured AI tutoring and study groups improve learning outcomes? Evidence from a randomized controlled trial

We investigate the effects of generative AI (Gen-AI) and peer collaboration on undergraduate macroeconomics student performance using a randomized controlled trial. Four course sections were randomly assigned to: (1) individual work without AI (control); (2) group work without AI; (3) individual work with Gen-AI; or (4) group work with Gen-AI. We developed a course-specific AI tutor, Macro Buddy, using course materials and a low-cost platform requiring no programming expertise. Student learning was assessed through three in-person exams, with confidence and perceptions measured through surveys. All treatment groups significantly outperformed the control group, with the combined Gen-AI and group-work condition producing the largest gains. The findings suggest that structured AI tutoring and peer collaboration complement one another in supporting learning.