Online learning often faces limitations due to unidirectional communication, where unresolved inquiries can disrupt learning flow and lead to student attrition. To address this issue, this study conducted a 2 × 2 factorial experiment (N = 52) using an AI avatar-based real-time Q&A system. The experiment manipulated two factors: response speed (fast vs. slow) and linguistic style (conversational vs. formal). The findings indicated that although the real-time Q&A function did not significantly improve mean learning achievement, it demonstrated a “stabilization effect” by reducing the variance in achievement scores. While the main effects of response speed and linguistic style were largely non-significant, a crossover interaction pattern with a medium effect size was observed for social presence and cognitive load. Specifically, social presence was notably higher in the conversational-fast and formal-slow conditions. Despite the statistical limitations of this small-scale pilot study, the findings provide preliminary evidence for the potential effectiveness of AI avatar-based interaction design in enhancing the online educational experience.