<![CDATA[Purpose – Grade-oriented learning in Indonesian secondary education hinders deep conceptual understanding, as reflected in declining PISA scores. This study examines the potential of AI-based formative assessment (ChatGPT-Padlet) to support improvement in students’ conceptual understanding and a shift in their orientation from a focus on grades to a focus on process. Design/methodology/approach – This single-cycle classroom intervention, drawing on classroom-action-research procedures, involved 33 tenth-grade students in Project-Based Learning supported by a layered feedback system (AI, peers, and teachers). Data were collected through pre- and post-tests, Likert-scale attitude surveys, peer assessments, and qualitative reflections. Findings – The intervention produced a statistically significant improvement in conceptual understanding scores (Δ = 9.09 points; Wilcoxon Signed-Rank Test, Z = -1.93, p = 0.045, r = 0.34; paired t-test reported as a robustness check, p = 0.026) and a non-significant descriptive decrease in competitive orientation (Δ = -0.3; Wilcoxon p = 0.32), while process focus remained stable in mean (M = 3.5, Wilcoxon p = 0.77) with increased variability. Qualitative analysis showed that 67% of students could articulate internet concepts, and 58% showed indications of transfer of knowledge to practical decision-making in their self-reported reflections. Research implications/limitations – The layered feedback ecosystem may offer a scalable approach to formative assessment in high teacher-student-ratio contexts, though this potential was not directly measured, and the single-cycle, short-duration design limits causal and longitudinal conclusions; further research requires longitudinal testing and more comprehensive deep learning metrics. Originality/value – This study offers an early empirical operationalization of process-oriented educational philosophy through an AI-peer-teacher triadic feedback model within the Indonesian Independent Curriculum context.]]>
