<![CDATA[This quasi-experimental study investigated the effectiveness of ChatGPT-integrated discovery learning on high school students' mathematical literacy in three-variable linear equation systems, while examining the moderating role of self-efficacy. Eighty tenth-grade students from a public high school in Indonesia were assigned to either an experimental group receiving ChatGPT-supported discovery learning instruction or a control group receiving conventional instruction, with mathematical literacy assessed through problem-solving tests and self-efficacy measured using validated questionnaires. Statistical analyses using non-parametric tests, including Wilcoxon Signed Rank Test, Mann-Whitney U Test, and Rank-Based ANCOVA, revealed that the experimental group achieved significantly higher post-test scores (M=79.25, SD=8.42) compared to the control group (M=65.30, SD=9.18), yielding a large effect size (d=1.24, p<.001) that substantially exceeded typical outcomes reported in meta-analyses of computer-assisted mathematics instruction. Contrary to established theoretical frameworks positing strong relationships between self-efficacy and mathematics achievement, this study found that self-efficacy neither significantly predicted mathematical literacy (F=3.171, p=.083, η²=.077) nor interacted with the learning model, suggesting that ChatGPT's pedagogical benefits operated uniformly across students with varying confidence levels. These findings suggest that ChatGPT functioned as adaptive conversational scaffolding that provided immediate formative feedback, facilitated metacognitive engagement through dialogic prompting, and generated multiple mathematical representations to support conceptual understanding. The null self-efficacy effect may indicate that robust AI scaffolding serves as an external compensatory mechanism that equalizes learning opportunities by reducing students' dependence on internal confidence judgments, thereby contributing novel theoretical insights into the conditional nature of motivational constructs in AI-mediated learning environments and offering practical implications for implementing conversational AI in mathematics education.]]>
Copyrights © 2026
