Answers are grounded in this paper and cite their sources.
The emergence of generative artificial intelligence (GAI) chatbots has raised concerns about the validity of unsupervised online assessments in higher education. This paper examines the impact of GAI on student performance by comparing outcomes in unsupervised online quizzes and supervised written exams before and after the widespread access to GAI, using data from a large-enrollment Finance course spanning eight cohorts (2021–2024) and more than 75,000 observations. By means of a differences-in-differences design, we find a substantial increase in performance on unsupervised assessments following GAI’s arrival, while no corresponding gains are observed in supervised exams, suggesting that the performance improvements reflect GAI-assistance rather than genuine learning. Consistent with GAI collapsing the search costs that previously made conceptual material hard to look up, the effect is driven almost entirely by theoretical questions, a stark reversal from the pre-GAI period, when students underperformed on such questions in unsupervised settings. Exploratory analyses show that the effect is similar across student ability groups, except for the highest-performing students, who exhibit only limited gains; that there is no evidence of gender differences; and that the effect grows across post-GAI cohorts, consistent with expanded adoption and improving tools.
No takes yet. Share an insight, caveat, or question.
Rodríguez-Longarela et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: