ChatGPT as cognitive scaffolding in 5E-based science instruction: Effects on middle school students’ problem-solving competence

Abstract

Problem-solving competence (PSC) is a central objective of contemporary science education; however, many students continue to experience difficulties when engaging in inquiry-based learning activities that require reasoning, reflection and solution refinement. This study investigated the effects of selective ChatGPT scaffolding embedded within the 5E instructional model—engage, explore, explain, elaborate, and evaluate—on middle school students’ PSC. A quasi-experimental design was conducted with 80 grade 9 students, including an experimental group receiving ChatGPT-supported 5E instruction and a control group receiving conventional 5E instruction. Unlike continuous AI integration, ChatGPT support was selectively provided during the explore and elaborate phases to facilitate inquiry reasoning, solution generation and reflective evaluation. PSC was assessed across three dimensions: problem understanding (PU), solution planning and implementation (SPI), and evaluation and improvement (EI). Baseline equivalence between groups was confirmed using independent-samples t-tests. Analysis of covariance results revealed a significant positive effect of the intervention on overall PSC, F (1, 77) = 6.080, p = .0159, partial η² = .073. Significant improvement was also observed for EI, F (1, 77) = 4.901, p = .0298, partial η² = .060, while SPI showed a positive trend, F (1, 77) = 2.953, p = .0898. No significant effect was observed on PU. The findings suggest that selective ChatGPT scaffolding is particularly effective in supporting reflective reasoning, evidence evaluation, and solution refinement in inquiry-based science learning environments.