Encouraging the use of generative AI for closed-book test preparation: effects on student assessment performance and perceptions

Generative artificial intelligence (GenAI) is increasingly being used by university students to support learning. However, evidence from authentic classroom settings examining whether GenAI-assisted revision influences closed-book assessment performance remains limited. This study evaluated the educational use of Microsoft Copilot in a third-year undergraduate immunology course. Student performance was compared between the 2025 and 2026 cohorts using comparable closed-book assessments of identical format and similar difficulty. Before Test 2 in 2026, students were encouraged to use Copilot during structured revision activities. Following Test 2, students completed an anonymous survey examining their perceptions of using Copilot to support learning. No statistically significant differences were observed in Test 1 or Test 2 marks between cohorts. Although a higher proportion of students achieved higher grades in 2026, a similar trend was evident before the intervention. Survey results showed that 78% of students agreed that GenAI could support learning of complex scientific concepts when used appropriately. However, only 53% reported improved understanding of immunology concepts, 39% reported improved knowledge recall, and 71% expressed concerns regarding inaccurate or misleading Copilot-generated information. These findings suggest that although students recognise the educational value of GenAI, encouraging its use during revision alone may not improve subsequent closed-book assessment performance.