Background: With the development of artificial intelligence (AI) technology, generative AI has been widely used in the field of education and represents a groundbreaking shift in overcoming the constraints of time and space within educational activities. However, previous literature has not paid enough attention to AI-involved teaching patterns, and it is necessary to evaluate the effects of this learning pattern. Objectives: Based on the social presence theory and the community of inquiry model, the main purpose of this study is to evaluate whether and how interaction frequency with chatbots (IFC) affects people's learning autonomy (LA) under two preferences: knowledge acquisition and virtual companionship, and whether social presence (SP) plays a mediating role. Methods: The 1-year longitudinal study was designed to be conducted from May 2022 to May 2023 and included three rounds of surveys of 1155 undergraduate students on their use of robots for learning. Results and Conclusions: For learners preferring virtual companionship, no direct correlation was found between IFC and LA. However, SP acted as a mediating factor, enhancing LA through increased chatbot interactions. This suggests that while direct interactions may not directly influence LA, the resulting SP can foster it. Conversely, for learners favouring knowledge acquisition, higher IFC negatively impacted both SP and LA. Despite this, a strong sense of SP consistently correlated positively with LA, indicating it could offset some negative effects of frequent chatbot use.