Faculty-precepted AI-generated clinical cases in preclinical biochemistry: a mixed-methods study informed by Gagné's instructional design and interpreted through Bourdieu's theory of practice

Background Generative artificial intelligence (AI) and large language models are increasingly permeating medical education, yet their integration into foundational biomedical sciences remains inadequately theorised and inconsistently evaluated. Biochemistry instruction, characterised by early curricular placement and perceived remoteness from clinical application, presents a compelling context for investigating whether AI-generated clinical cases can bridge the divide between theory and practice. Objective To evaluate the educational impact of AI-generated clinical cases embedded within a spiral undergraduate biochemistry curriculum, examining learner perceptions, academic performance, and alignment with international competency frameworks, interpreted through Gagné's instructional model and Bourdieu's theory of practice. Methods We conducted a convergent mixed-methods study within a first-year undergraduate biochemistry course ( n = 81). The intervention comprised student-generated, faculty-verified clinical vignettes produced using large language models, delivered through tutorials structured according to Gagné's Nine Events of Instruction. Quantitative data included a purpose-designed questionnaire and comparative grade analysis against a historical cohort. Qualitative data from three performance-stratified focus groups were analysed thematically using Braun and Clarke's framework, with quality appraisal against Standards for Reporting Qualitative ResearchStandards for Reporting Qualitative Research standards. Competency alignment was mapped to six international frameworks. Results Learners reported favourable perceptions across cognitive understanding, clinical relevance, and peer collaboration. A thematic analysis revealed initial apprehension with subsequent adaptation, accuracy concerns necessitating verification literacy, deepened conceptual understanding alongside cognitive load considerations, and implementation challenges requiring faculty scaffolding. The intervention cohort showed enhanced grade distributions, with upward shifts in higher achievement bands. Interpreted through Bourdieu, the findings may be read as reconfiguring learner habitus towards active knowledge construction, introducing technological and verification capital, and intensifying the negotiation of epistemic authority. Competency mapping indicated alignment with knowledge application, communication, and scholarly practice domains. Comparative grade findings are reported descriptively (as proportions and percentage-point differences) and were not inferentially tested; they should be interpreted as exploratory. These outcomes reflect a multi-component instructional design (student-generated, collaborative, AI-assisted, and faculty-verified) and are not attributable to generative AI alone. Conclusions AI-generated clinical cases represent a promising pedagogical innovation for contextualising foundational biochemistry when embedded within explicit instructional frameworks and verification structures ( Datasheet 1 ). Differential benefits across performance strata, platform-related accuracy variability, and equity considerations regarding tool access warrant deliberate curricular calibration.