Key points are not available for this paper at this time.
Digital storytelling is widely used in education as a multimodal design practice that foregrounds creator agency and audience awareness. Generative AI (GenAI) expands the speed and range of story production but also raises questions about how educators retain pedagogical control over generated content. This exploratory qualitative study used cross-case workflow analysis to examine how 27 pre-service early childhood educators in Hong Kong designed GenAI-mediated mathematics storybooks. Data included video-recorded process presentations, records of interactions with ChatGPT (GPT-5) and Krea 2, and completed storybooks. Analysis traced participant decisions across planning, generation, curation, and integration, focusing on GenAI-related affordance uptake, recurring workflow patterns, and tensions among technical innovation, ethical accountability, and cultural impact. Findings show that participants commonly used GenAI to steer generation through pedagogical constraints; expand narrative and visual alternatives; and select, reject, or revise generated materials. Two recurring process patterns were identified: constraint-guided planning-to-generation, in which pedagogical requirements were established early and used to guide subsequent decisions, and exploration-to-selection, in which participants generated and compared alternatives before progressively stabilizing a preferred direction. Across both patterns, tensions were particularly evident during curation and integration, when participants balanced efficiency with sustained oversight, creative variation with pedagogical coherence, and generic outputs with local contextual fit. The findings show that GenAI-mediated storybook creation is not simply a process of generating text and images, but a pedagogical workflow in which pre-service teachers exercise agency through the critical selection, revision, and integration of AI-generated materials. Implications are discussed for teacher education, AI literacy, and the design of scalable support for teachers’ use of GenAI.
Chan et al. (Tue,) studied this question.