New Perspectives in Science Education

Students learning chemistry often struggle to construct coherent explanations that connect observable phenomena with particle-level mechanisms and symbolic representations. While Socratic questioning may effectively support such reasoning, its sustained use in secondary classrooms is frequently constrained by time, class size, and instructional demands. Advances in generative artificial intelligence (AI) offer potential support for dialogic inquiry. Yet, how such tools can be used in instructionally disciplined and pedagogically meaningful ways remains underexplored in classroom settings.

This pilot study explores how a prompt-engineered Socratic-style AI chatbot can be integrated into inquiry-based chemistry instruction to support students’ chemical reasoning. The AI was deliberately designed to function as a questioning scaffold rather than an answer-providing tutor, posing probing and reflective questions while withholding explanations and evaluative feedback. The intervention was implemented across two consecutive class sessions within a 5E inquiry cycle (Engage–Explore–Explain–Elaborate–Evaluate) in an upper-secondary chemistry classroom, focusing on periodic trends and their mechanistic explanations, including atomic size, ionization energy, electronegativity, and selected exceptions.

Data sources include pre- and post-inquiry diagnostic assessments targeting the quality of students’ chemical reasonings, AI–student dialogue logs, written inquiry artifacts, and follow-up interviews with a purposively selected anchor group representing different reasoning trajectories. Analysis focuses on exploring changes in the structure and coherence of students’ reasonings, particularly shifts from descriptive rule-based reasoning toward more mechanistic accounts. It also focuses on identifying how Socratic AI prompts mediate moments of cognitive conflict, revision, and reflection during inquiry.

By examining the feasibility and pedagogical affordances of constrained Socratic AI use in authentic classroom contexts, this pilot study aims to identify interaction patterns, design considerations, and limitations that can inform the development of larger-scale investigations into AI-supported inquiry learning in science education.