Abstract
Teacher-researchers who study their own classrooms increasingly face more student writing than they can read closely, and generative artificial intelligence (AI) promises relief. Yet automating the analysis risks surrendering the interpretive judgment that makes it research. Drawing on a study of Mexican upper-secondary students’ mathematical myths, we describe a six-phase model that integrates a large language model into conventional content analysis while keeping interpretation in human hands. Across immersion, initial labeling, consolidation, category construction, codebook building, and corpus-wide application, we specify what the AI does and what the teacher-researcher decides. Three simple tools—a color-coded evidence sheet, a codebook with explicit rules, and marginal notes—keep the teacher-researcher in charge. The organizing principle is deliberately blunt: the AI counts; we interpret.
Keywords
content analysis, artificial intelligence, large language models, teacher research, mathematics education
How to Cite
Nava Guzmán, C. & Aguilar, M., (2026) “The AI Counts, We Interpret: A Six-Phase Model for Partnering with Language Models in the Analysis of Students’ Mathematical Writing”, Ohio Journal of School Mathematics 104(1): 7, 71--83. doi: https://doi.org/10.18061/ojsm.7676
