Addressing Bloom’s 2 Sigma Problem, this study investigates how a GPT-based differentiated 1:1 self-paced coach can approximate the benefits of one-to-one tutoring in middle school chemistry. Chemistry learning poses unique challenges due to its reliance on abstract representations and persistent misconceptions—especially in balancing chemical equations, where students often resort to trial-and-error strategies without conceptual understanding. Traditional instruction frequently lacks the individualized scaffolding needed to address these issues. To tackle this, we integrated the 5E Learning Cycle (Engage, Explore, Explain, Elaborate, Evaluate) as the core pedagogical framework, with GPT prompts carefully designed following the 5S Prompting Principles (Set the Scene, Be Specific, Simplify Language, Structure Output, Share Feedback). Sixty eighth-grade students participated in a quasi-experimental study. The experimental group received AI-assisted coaching, while the control group received traditional instruction. Results revealed significantly higher post-test scores and greater consistency for the experimental group, with a large effect size. Lag sequential analysis further demonstrated that GPT-student interactions aligned with 5E inquiry patterns. This study offers a replicable model for no-code generative AI integration into chemistry instruction and highlights the potential of structured LLM applications to support scalable, personalized, and conceptually rich learning experiences.

Authors:
Teng-Chi Lin, National Taichung University of Education, Taiwan
Cheng-Hsuan Li, National Taichung University of Education, Taiwan

About the Presenter(s)
Mr. Teng-Chi Lin, a high school chemistry teacher and Ph.D. candidate at National Taichung University of Education, focuses on chemistry education and digital learning. His current project applies generative AI to enhance science learning.

See this presentation on the full schedule – Saturday Schedule

Bookmark