Abstract Background Learning‐by‐teaching is a generative learning strategy in which students are told they will have to teach what they are learning to others. Although learning‐by‐teaching has been shown to be effective in some cases, few studies have established guidelines for how to optimize the benefits of learning‐by‐teaching as a generative learning strategy from a social presence perspective. Objectives This study seeks to clarify the learning‐by‐teaching hypothesis and to pinpoint the optimal level of social presence during learning‐by‐teaching that is most conducive to learning. Methods In Experiment 1, college students received a lesson with instructions that afterwards they would explain the material to others by making a video, explain the material aloud to themselves, or restudy the material. In Experiment 2, college students viewed a multimedia lesson with instructions that afterwards they would explain the materials by making a video, explain to an onscreen student, or explain to a student in person. Results and Conclusions Teaching by making a video was better than restudying, self‐explaining, and teaching face‐to‐face or online. Teaching quality was better in video teaching than self‐explaining and face‐to‐face or online teaching. Teaching by making a video is ideal because it primes generative processing while minimizing extraneous processing. Implications This study is the first to manipulate different levels of social presence of oral teaching to determine the optimal form of learning‐by‐teaching, which preliminarily clarifies generative learning and social presence theory and has implications for both empirical and theoretical research.