This study aims to investigate a mathematical-metacognitive discourse model with IMPROVE metacognitive self-question prompts (What, How/When, and Why) in order to promote a deeper level of discourse in mathematics classes. The study involved 32 math teachers and 824 fifth-grade students (from 32 public schools) from one district in Israel. Participants were assigned to a randomized, controlled quasi-experimental trial, including an experimental group exposed to mathematical-metacognitive discourse, focusing on metacognitive skills (i.e., planning, monitoring, and reflection) used to arrive at solutions, and a control group exposed to mathematical discourse, focusing on cognitive knowledge (i.e., declarative, procedural, and explanatory). An in-depth analysis was conducted, using a mixed approach (quantitative and qualitative), of a recorded video teacher-student class discourse lesson at the end of the study in each class (n = 32) as well as case studies of two teachers (one from each group) focusing on discourse pattern and their talk moves in class. The findings indicated that the experimental group exhibited more conceptual verbalization related to planning and reflection processes, whereas the control group scored the highest with regard to procedural knowledge. As a transfer measure from the intervention program at the end of the study, students in the experimental group outperformed the control group in terms of problem-solving and sensemaking performance. This study contributes further to understanding discourse by introducing the concept of mathematical-metacognitive discourse competence. Moreover, the study has practical implications for the promotion of discourse in mathematical lessons.