<![CDATA[Generative artificial intelligence offers new opportunities to scaffold students’ mathematical reasoning, yet rigorous evidence of its impact on secondary students’ problem-solving remains limited. This study examined whether ChatGPT-driven adaptive learning improves eighth-grade students’ problem-solving performance on systems of linear equations in two variables (SPLDV) compared with conventional instruction. A pre–post-test control-group design was implemented with 47 eighth-grade students in Parepare, Indonesia (experimental n = 24; control n = 23) during the 2024/2025 academic year. The experimental group used ChatGPT as an adaptive tutor aligned with Polya’s stages (understand, plan, execute, look back) through guided prompts, hints, and feedback. In contrast, the control group received a lecture and practice. Students completed a six-item contextual SPLDV test scored with a Polya-based rubric. Between-group differences were tested on post-test scores and normalized gains after verifying normality and homogeneity assumptions. The experimental group achieved higher post-test performance (M = 68.33) than the control group (M = 59.57), with a significant difference (p = 0.019; η² = 0.117). Learning gains were also larger in the experimental group (mean N-gain = 0.34, medium) than in the control group (0.21, low; p = 0.001; η² = 0.372). Indicator-level patterns suggested the greatest improvements in understanding the problem and carrying out the plan, whereas devising a plan remained the most challenging stage in both groups. These findings indicate that ChatGPT-based adaptive scaffolding can enhance students’ mathematical problem-solving on SPLDV, but explicit teacher-guided routines are needed to strengthen strategic planning and the critical evaluation of AI outputs.]]>
Copyrights © 2026
