Bridging the gap: health professions students’ preferences for AI curriculum integration and perceived barriers

Background Integrating artificial intelligence (AI) into health professions curricula poses significant pedagogical challenges. Grounded in the Technology Acceptance Model (TAM), this study examined how undergraduate health professions students perceive AI educational modalities, evaluated tool effectiveness, and identified barriers to acceptance. Methods Between January and February 2026, 713 students from three institutions in eastern China completed a validated online survey. Stratified random sampling ensured representation across academic years (1–5) and disciplines (clinical medicine, medical laboratory science, and allied health programs). The instrument assessed preferences for four instructional formats and three AI tool categories. Analyses included descriptive statistics, Cochran’s Q tests, paired t-tests, mixed-effects logistic regression, and thematic analysis of 412 open-ended responses. Results Integration into existing coursework was the dominant preference across all tool types (41.65–44.74%), whereas standalone modules attracted limited support (16.69–21.74%). Mixed-effects logistic regression showed that senior students (Years 4–5: OR 1.79, 95% CI 1.23–2.60, p = .002) and those reporting high perceived usefulness (OR 2.44, 95% CI 1.71–3.48, p < .001) and ease of use (OR 1.51, 95% CI 1.04–2.19, p = .031) were significantly more likely to favour the integrated approach. A usability paradox was observed: specialized medical AI tools were rated marginally more useful than general large language models (mean 4.43 vs. 4.12; Cohen’s d = 0.24), yet less easy to use (mean 3.89 vs. 4.35; Cohen’s d = 0.31). Given the limited clinical exposure reported by many respondents, these ease-of-use ratings likely reflect perceived rather than experienced usability. Key adoption barriers were data privacy concerns (30.0%) and ethical uncertainty (21.9%). Participants valued clinical applicability over breadth of coverage (4.67 vs. 4.38, p < .001). Conclusions Findings support moving away from isolated AI courses toward clinically embedded training. Curricular reform should prioritise hands-on relevance and ethical governance while addressing the usability barriers of specialised platforms, thereby preparing future health professionals to direct AI applications critically rather than rely on them passively.