Leveling Up or Leveling Down? The Impact of Generative AI on Student Performance in Business Schools

This study investigates how generative AI (GenAI) access impacts student performance in ill-defined, time-pressured business school exams. Through an embedded mixed-methods design combining an experimental study with qualitative interviews, we identify an equalizing effect: low performers improve while high performers decline, resulting in performance convergence. Our qualitative analysis reveals the mechanism driving this convergence—GenAI-induced cognitive load inversion. Low performers experience cognitive load relief by copying chatbot output, thus bypassing the analytical work the task requires. High performers experience cognitive load amplification, struggling to process voluminous output under time pressure, disrupting their analytical processes. We argue that task structure shapes GenAI’s effects in time-constrained situations: the ill-defined nature of our task elicits different cognitive challenges compared to well-defined tasks of prior research, helping reconcile mixed findings on GenAI’s democratizing effects. Furthermore, the findings reveal how traditional assessments fail when GenAI masks performance differences.