Lesson planning with ChatGPT for inquiry-based biology instruction – A(I) roll of the dice?

This study investigates the capability of ChatGPT-4o to generate high-quality inquiry-based science lesson plans, that is, aligning all elements of a written lesson plan to students’ learning about procedural and epistemic aspects of science instead of gaining subject matter knowledge. Using an exploratory sequential mixed methods design, we analysed <i>N</i> = 60 biology lesson plans generated by the research team across four key topics (cell biology, genetics/evolution, human biology, ecology) and five scientific inquiry practices (microscopy, observing, experimenting, modelling, reflecting the nature of science) from the German curriculum . First, lesson plans were quantitatively evaluated using a modified version of Großmann and Krügers’ [(2024). Assessing the quality of science teachers' lesson plans: Evaluation and application of a novel instrument. <i>Science Education, 108</i>(1), 153–189. https://doi.org/10.1002/sce.v108.1] scoring rubric, which assessed ten quality criteria with substantial interrater agreement (Cohen's <i>κ</i> = .67). Second, based on the score distribution, we conducted qualitative analyses to identify strengths and weaknesses in ChatGPT-generated inquiry-based lesson plans. Results revealed considerable variation in lesson plan quality, with <i>n</i> = 22 lesson plans achieving less than 50% of the maximum possible score and only <i>n</i> = 5 lesson plans reaching 75% or higher. While the lesson plans demonstrated particular strengths in content accuracy and learning outcome alignment, they exhibited significant weaknesses in addressing students’ inquiry-related conceptions and maintaining consistent focus on scientific inquiry. While ChatGPT-4o successfully generated some high-quality lessons plans for inquiry-based instruction, many lesson plans shifted inappropriately towards subject matter knowledge acquisition rather than scientific inquiry processes. This discrepancy between our initial prompts and the resulting lesson plans indicates that while large language models like ChatGPT-4o demonstrate potential as preliminary planning tools, they necessitate thorough evaluation and substantive modification by teachers possessing robust pedagogical content knowledge. The findings emphasise that the advent of generative artificial intelligence does not diminish the importance of professional knowledge in teacher education but rather transforms how this knowledge is applied in lesson planning processes.