ABSTRACT Technological advances in video‐based education practices have led to the creation of pedagogical agents (PA), and an important question now is how to optimize their visual presence. This study explored how different PA presence forms in instructional videos influence children's learning. Four videos were created, each featuring a different type of PA presence: continuous presence of a realistic or a virtual panda, alternative or interactive presence of human and panda PAs. In the experiment, 112 third‐grade students were randomly assigned to watch one of the videos. Results showed that the interactive presence video led to the highest scores in retention, transfer, learning motivation, social presence, and germane cognitive load. Social presence mediated the influence of interactive presence on retention, whereas social presence and germane cognitive load mediated the influence on transfer. The findings confirm and better clarify the benefits of introducing interactive PA presence design in children's video learning.