Chatbots have shown great potential for language learning. However, previous studies have reported mixed results on the efficiency of chatbot-assisted language learning (CALL). This study integrated the results of previous experimental studies on CALL by using meta-analysis to explore its effectiveness. A total of 61 samples from 18 studies were examined. The results showed that CALL had a moderate average effect (g = 0.527). In addition, nine potential moderating variables (educational level, target language, language domain, learning outcome, instruction duration, chatbot interface, chatbot development, task dominance, and interaction way) were identified and discussed. The results of this study provided insights into the use and design of chatbots for language learning.