AI chatbots developed by users or commercialized for English education have been recently welcomed for providing interaction opportunities to English learners. In this context, the present study, which focused on the need for pedagogically appropriate responses of chatbots to erroneous utterances, examined the response patterns to errors of two commercial chatbots (i.e., AI Pengtalk and Speaking Class) currently utilized in numerous English lessons. To this end, the researchers intentionally uttered pronunciation, vocabulary, and grammar errors reported as being frequently committed mistakes by young Korean learners of English to the chatbots. Then, the researchers analyzed the chatbots' responses to the erroneous utterances, primarily focused on error recognition, meaning negotiation, and corrective feedback. As a result, the chatbots responded appropriately to some errors in pronunciation of interdental fricative and vocabulary omission as it judged the errors as low-level achievement and provided meaning negotiation. However, the chatbots wrongly judged most other pronunciation, vocabulary, and grammar errors as correct utterances; hence, no appropriate interactional responses were given to the errors. Considering the response patterns of the chatbots to the errors as inappropriate for English learners' interlanguage development, this study suggests data augmentation of learners' erroneous utterances for the chatbots to provide more meaning negotiation and corrective feedback.