<p>As artificial intelligence (AI) has become central to advancing educational modernization, AI teaching assistants have also emerged. These innovative instructional tools provide knowledge services while also enhancing learners’ metacognitive abilities through human-AI interaction. This study collected quasi-experimental data from a university to examine the effects of AI teaching assistants on students’ learning outcomes. By incorporating questioning strategies into the conceptual framework, this study shifts the focus from whether students use AI teaching assistants to how they use them, thereby uncovering the mechanisms behind the differences in learning outcomes. The results showed that the use of AI teaching assistants significantly improved learning outcomes. Students who used AI teaching assistants scored an average of 9.09 points higher than those who did not, with a 36.04% reduction in grade variability. This effect was notable among low- and mid-performing students. Furthermore, when using AI teaching assistants, the knowledge-reflective questioning strategy had a significantly positive effect on learning outcomes compared to the copy-pasting strategy. Finally, questioning quality was more important than frequency. As the frequency of AI teaching assistant usage increased, the knowledge-reflective questioning strategy consistently generated positive effects on learning outcomes, exhibiting a trend of diminishing marginal returns. However, the AI usage frequency in the copy-pasting strategy group had a non-significant and slightly negative effect on learning outcomes. Overall, this study provides new empirical evidence in educational technology and learning sciences, offering practical implications for optimizing the interaction design of AI teaching assistant systems and refining teaching management strategies.</p>