Decoding medical students’ attitudes toward ChatGPT: Psychometric evaluation of the Persian version of the attitudes toward ChatGPT questionnaire

The rapid emergence of generative AI like ChatGPT in medical education necessitates understanding student attitudes for its effective integration. This study aimed to translate, culturally adapt, and validate the Persian version of the Attitudes toward ChatGPT (CAS) scale to address this need. A cross-sectional methodological study was conducted with 421 undergraduate medical students from Iranian medical universities. The 21-item CAS developed by Yu et al. (2024) was translated and adapted following Beaton’s cross-cultural validation framework. Data were analyzed using exploratory (EFA) and confirmatory factor analyses (CFA) with maximum likelihood estimation. Internal consistency was examined using Cronbach’s α and McDonald’s ω; convergent and discriminant validity were assessed via composite reliability (CR), average variance extracted (AVE), and the Fornell–Larcker criterion. EFA supported the original five-factor structure—Tool, Tutor, Threat, Talk, and Trend—accounting for 44% of total variance (KMO = 0.81; RMSEA = 0.049). CFA confirmed acceptable model fit (χ 2 /df = 2.08; CFI = 0.922; TLI = 0.905; RMSEA = 0.051). Reliability was satisfactory across subscales (α = 0.63–0.82; ω = 0.67–0.84). Convergent validity was strong for Talk (AVE = 0.69) and Trend (AVE = 0.57), while discriminant validity was mostly supported, except for partial overlap between Tool and Tutor. One item (Q21) demonstrated a Heywood case and requires refinement. The Persian version of the CAS demonstrates preliminary psychometric support for assessing medical students’ attitudes toward ChatGPT, with an interpretable five-factor structure and acceptable overall model fit. However, item-level instability, suboptimal convergent validity for several subscales, and conceptual overlap between the Tool and Tutor dimensions indicate that the scale requires further refinement. Accordingly, the Persian CAS should be considered a developing measurement framework, suitable for exploratory research and construct-level investigation, rather than a finalized instrument for applied or high-stakes educational use.