Conversations with others are important in order to improve an individual’s mental state. Recently, LLM-based dialogue systems, such as ChatGPT, facilitate natural conversation and have the potential to improve mental state such as emotions or mood. Our aim is to develop a dialogue system that generates responses tailored to the user’s personality and physiological state. As part of this study, we set the research question “Does dialogue with ChatGPT, and the interplay of the extroversion between the user and ChatGPT, improve the user’s mental state?”. In order to verify this research question, we implemented a dialogue application, and conducted a total of three speech dialogue experiments using a ChatGPT API. Participants rated their mental state along seven emotion axes (e.g., Sad-Happy, Anxious-Relaxed), using a 15-point scale both before and after each dialogue. To set the extroversion level (low, average or high) for ChatGPT, we added information of ChatGPT’s extroversion into a prompt. Our experiments showed that the mean of seven emotion axes significantly improved after dialogues. Moreover, although not statistically significant, the largest differences between before and after dialogue were observed when both participants and ChatGPT had either a similarly low or high level of extroversion.