Collaborative Online International Learning (COIL) is a teaching method that connects students and teachers from different countries to collaborate on projects within their curriculum. Although numerous studies have examined COIL, none, to our knowledge, have addressed AI-assisted language learning in COIL interactions, particularly in Japanese higher education. This presentation explores the learning outcomes of using personalized oral tasks with ChatGPT during COIL. These tasks were tailored to each student’s specific learning needs to prepare them for successful COIL interactions. This study involved Japanese students participating in two COIL projects with American universities. AI-assisted learning occurred shortly after the first COIL project and before the second. To track students’ progress, transcripts from their practice conversations with ChatGPT were collected, along with recordings of their interactions during Zoom meetings. Discourse analysis was employed to interpret the data and understand how students used the language practiced with ChatGPT in their interactions. The findings indicate that students developed conversational strategies through practice with ChatGPT, which they then applied during their second COIL interactions. Additionally, the analysis of students’ reflections revealed their perceived benefits of AI-assisted learning for improving conversational skills. The implications of this approach for real-life communication in virtual settings, are discussed.