Designing Diagnosis-Informed Microlearning with AI-Feedback for Stoichiometry Learning

Stoichiometry is conceptually challenging because students may experience misconceptions that are not adequately revealed through conventional achievement tests. This study aimed to design and conduct a limited proof-of-concept of diagnosis-informed microlearning with AI-feedback to support students’ conceptual understanding of stoichiometry. The study involved 68 Grade X students from two senior high schools and five expert validators. Thirty diagnostic items were validated, with 24 meeting the predetermined content-validity criteria, from which 18 items were selected to represent six stoichiometry subtopics. A separate 30-item conceptual understanding test yielded 24 valid items, from which 20 items were selected for limited implementation. The initial diagnostic assessment was administered to 34 students, while the revised learning website was implemented with another 34 students. The diagnostic results showed the highest proportion of conceptual problems in limiting reagents (67.7%). Following the limited implementation, 52.94% of students were classified as having moderate-to-high conceptual understanding. Students’ responses to the AI-feedback website were generally positive, with positive responses exceeding 75% across all indicators. The findings provide initial evidence of the potential of diagnosis-informed AI-feedback for formative stoichiometry learning, while further controlled studies are required to establish its effectiveness.