Abstract:

The rapid diffusion of generative artificial intelligence has created a new question for tertiary teaching: can an AI tutor complement student learning, or does it mainly substitute for it? In this paper, we evaluate the introduction of an AI tutor in an introductory economics course and examine its relationship with student learning and student experience. The AI tutor was implemented as a set of custom GPTs aligned to the paper’s four tests, with each custom GPT covering three course topics. The tutor was grounded in topic-specific learning objectives, lecture slides, and lecture transcripts, and was designed to support learning rather than provide direct assessment solutions: it had no access to test or tutorial solutions and was instructed not to generate certain problem types. The study uses data from two trimesters in 2025, combining student demographics, test scores, adjusted overall grades, and data on whether and how often students accessed the tutor.

Because access was not randomised, the analysis is explicitly associational rather than causal. The empirical strategy therefore combines cross-sectional OLS models of overall course performance with panel regressions at the test level that treat each student as their own control. Models examine both the extensive margin of tutor use and the intensive margin of frequency of use, with and without controls for demographics, engagement, prior economics knowledge, and academic ability measured by GPA in papers excluding introductory economics.

Across specifications, students who used the AI tutor, and among users those who used it more intensively, tended to perform better overall and on individual tests than students who used the AI tutor less or not at all. These positive associations appear stronger for lower-GPA students, with preliminary evidence that they may also be somewhat larger for students with lower prior economics knowledge, suggesting a possible catch-up effect. There is also weak evidence of spillovers onto performance in other papers, consistent with the possibility that some students reallocated study time. Just as strikingly, student feedback indicates very strong acceptance of the AI tutor as a learning support tool.

Keywords:

Artificial intelligence, AI tutor, Economics.