AI voice-based chatbots (AVCs) are able to act as promising learning assistants, enhancing speaking proficiency as conversation partners. However, little is known about how learners and AVCs actually interact with each other in such conversations. Drawing on (Long, Handbook of second language acquisition, Academic Press, 1996) Interaction Hypothesis, this study investigates how Vietnamese university students interact orally with ChatGPT, an AVC, in speaking classes. Through an exploratory, mixed-methods research design using both reflective surveys and screen recordings, the chapter examines the types of linguistic adjustments ChatGPT made during chatbot-human interactions, the ways both the students and ChatGPT negotiated for meaning, and how the students modified their language outputs during conversations with ChatGPT. The results reveal that, overall, students have a positive attitude towards integrating ChatGPT in practicing speaking skills. The results also show that both students and ChatGPT have adjusted their language to negotiate for meaning, such as adjusting speech rate, asking for clarification and repetition, using paraphrases, and elaboration. These results highlight the engagement of chatbot-human interactions and their benefits for students’ speaking practice and performance. The chapter concludes by discussing the implications of these findings for using AVCs as an effective conversation partner to boost students’ speaking skills and suggests pathways for further research.