AI-based causal evaluation of teacher’s opening lessons: a multidimensional study

The opening phase of a lesson can strongly influence students’ readiness to learn, yet teachers often receive limited, delayed feedback on the quality of this phase. This study examines which opening-lesson features are most strongly linked to higher-quality lesson introductions in chemistry classrooms. We analysed 200 classroom videos of Indonesian chemistry teachers. Video audio was transcribed and then scored using a structured rubric covering three features: (1) attention-building, (2) teacher–student interaction, and (3) use of instructional media. In this study, “lesson outcomes” refers to the rubric-based quality score of the opening phase (i.e., the overall opening-lesson score), not student achievement scores. Descriptively, teachers scored highest on interaction (mean = 3.44), while media use showed greater variation (mean = 3.04). Using an observational design with causal-inference analyses to reduce selection bias, we found that stronger attention-building and richer teacher–student interaction were associated with higher opening-lesson quality. Media use appeared to strengthen these benefits, partly by supporting interaction (media → interaction → overall opening score). Sensitivity checks suggested the main findings were stable under moderate unobserved bias. Overall, the results highlight that the most effective lesson introductions combine clear attention cues, meaningful interaction, and purposeful media use. Teachers with lower baseline performance showed the greatest improvement when openings included clear learning objectives, suggesting a need for targeted, personalised professional development. This AI-supported evaluation approach offers a scalable way to provide more consistent feedback on lesson introductions for teacher development.