Search for a command to run...
Purpose This study aims to examine how artificial intelligence (AI)-guided versus human-guided museum educational tours shape teenagers’ long-term memory formation, focusing on cognitive and socio-affective pathways such as vividness, storytelling, parasocial interaction, perceived expertise, self-efficacy and semantic integration. Design/methodology/approach A field-based quasi-experiment was conducted with 414 teenagers (AI-guided = 229; human-guided = 185), followed by a survey 28-days postvisit. Hypotheses were tested using partial least squares structural equation modeling and multigroup analysis. Findings Semantic integration is the universal core mechanism for long-term memory. The key theoretical contribution lies in revealing medium-contingent antecedent pathways: storytelling is a critical driver of perceived expertise only in AI-guided tours, whereas self-efficacy is the consistent robust driver across both. This establishes a novel medium-contingent cognitive model for technology-enhanced learning. Research limitations/implications Limitations include the quasi-experimental design, single-museum context and partial reliance on self-reports. Future research should use randomization, multisite sampling and longitudinal and/or multimodal measures of memory and learning processes. Practical implications Museums should prioritize layered, educational storytelling in AI-guided tours to strengthen perceived expertise and downstream semantic integration. Human guides should emphasize efficacy-building interactions and adaptive scaffolding. Educators can design pre- and post-visit activities that explicitly prompt semantic integration. Social implications As museums scale AI-based interpretation, attention is needed to digital equity, ethical governance and inclusive design so that technology-enhanced learning benefits diverse teenage visitors. Originality/value This study pioneers a medium-contingent cognitive model, integrating levels of processing theory with human–AI interaction constructs. This provides novel insights for designing technology-enhanced learning environments.