Review of 53 Studies: LLM Lesson Plans Well-Structured but Lack Pedagogical Depth

A research team from the University of Thessaly in Greece published a narrative review in Frontiers of Digital Education analyzing 53 recent publications on large language model-generated lesson plans. The review finds that tools such as ChatGPT can produce clearly structured plans but often lack contextual awareness, differentiation, and pedagogical depth for direct classroom use. It also reports that a plan designed for one hour required at least three hours in practice, and output quality varied within identical prompt groups.

This review provides an evidence boundary for teachers and product developers on current LLM lesson-planning capabilities: the tools can assist with initial organization, but teachers retain final responsibility for adaptation and application. For AI education products, differentiation, time planning, and output consistency are clear improvement targets; for schools, teacher AI literacy training and clear usage policies are prerequisites. The review also notes that most studies rely on teachers' subjective opinions, with limited classroom application and direct comparisons between teacher-created and LLM-generated plans, so practical extrapolation requires caution. Teachers can use LLMs for initial lesson-plan frameworks but need substantial time for differentiation and pedagogical depth adjustments. AI education product developers should prioritize output consistency, special education adaptation, and classroom time estimation.