Aiming to alleviate anxiety, this study introduces Bartender’s Advice, a serious game based on Generative Agents. It integrates mutual aid therapy and cognitive behavior therapy through interactive tasks between players and non-player characters (NPCs) facing anxiety, promoting self-reflection and positive expression tailored to anxiety issues. This design strategy enhances players’ self-efficacy and facilitates autonomous cognitive intervention. On the technical side, Dialog Based Evolving Generative Agents (DBEA), developed using large language models, improve player engagement by providing customized, iterative, human-like interactions. The experiment used the Spielberger Anxiety Inventory (SAI) scale to measure changes in anxiety levels before and after gameplay. Results indicated an effective reduction in users’ anxiety states, with qualitative feedback highlighting positive gaming experiences and perceived therapeutic benefits.