Abstract / Summary
Background: Beyond technical feasibility of Artificial Intelligence (AI)-generated multiple-choice questions (MCQs), their educational value in assessment remains unclear. This study aims to evaluate whether students can distinguish between AI-MCQs and National Licensing Exam (NLE) questions in an exam setting and if they align with the curriculum.
Methods: In this cross-sectional study 119 year five medical students completed a Family Medicine MCQ tablet-based exam. Participants answered 30 AI-generated and 30 NLE MCQs. AI questions were generated from digital learning materials using ChatGPT-4o and Gemini 1.5 Pro. Experts accepted 82% of the questions, with minor edits to retained items and removing those needing major changes. During the exam, students were asked about the question source and whether each item aligned with the course curriculum. Statistical analyses were obtained using Jamovi 2.3.28.0.
Results: No significant difference in correct attribution of AI or NLE MCQs (t(29.6) = -1.24, p = .225; t(29.6) = 1.18, p = .246) was observed. No significant correlation was found between item difficulty and recognition (τ_b: p = .534). Distractor distributions did not differ across ChatGPT, Google Gemini and NLE (χ2(2) = 2.61, p = .271, U p > .05 for all comparisons). Item difficulty did not differ significantly between AI and NLE items; however, exploratory source-specific analysis showed an overall difference among ChatGPT, Google Gemini, and NLE items (χ²(2) = 6.71, p = .035), with a difference between Google Gemini and NLE items (p = .028). Student-perceived curricular alignment did not differ significantly between AI and NLE or among ChatGPT, Google Gemini, and NLE.
Conclusions: Students' recognition did not differ between AI-generated MCQs and NLE MCQs. Easier MCQs are generally perceived as more aligned with the curriculum. AI-MCQs may provide a feasible approach to item drafting within a structured human-review process.