The use of evidence-based practice can be regarded as the gold standard in education. A steadily adopted educational practice is non-interactive teaching, in which students explain the previously learned contents to a fictitious peer. Although recent laboratory studies documented the benefits of non-interactive teaching, field-oriented evidence is scarce regarding different implementations, contents, student populations, and domains. In this field study, we applied the ManyClasses approach to test the generalizability of non-interactive teaching in diverse contexts to explore external boundary conditions (domain, school type, grading, medium and timing of the learning task) and student-related boundary conditions (age, gender, native language). We examined the effect of non-interactive teaching in k = 20 different teaching units (each 2 lessons) across various school types and subjects in which school students (N = 191) were randomly assigned to sequences of non-interactive teaching and retrieval practice (within-participants design). We did not obtain a main effect of non-interactive teaching; but the domain, school type, and grading moderated the effectiveness of teaching. The findings demonstrated the feasibility of the ManyClasses approach and highlighted the need of considering contextual variables to make generalizable recommendations for educational practice at the same time.