Loneliness, social isolation, and anxiety affect millions of people across the world. Communication technologies, including artificial intelligence (AI) chatbots, can potentially offer support to those experiencing mental health challenges by providing companionship and support. Specifically, social AI can mimic human interaction, which may help alleviate loneliness and anxiety through person-centered messaging. Despite growing AI usage, there is limited research on the effectiveness of specific message types in this context. Thus, this study employed a 2 (person-centered message: high vs. low) × 2 (context: loneliness vs. anxiety) between-subjects design to test how different supportive messages from social AI chatbots impact subsequent outcomes. Results revealed that high person-centered messages are associated with increased emotional validation. Furthermore, the quality of social support and interpersonal warmth (IW) mediated the relationship between high person-centered messages and emotional validation. Finally, the mediation effect between high person-centered messages and emotional validation via the quality of emotional support was moderated by social presence, but not the mediation effect between high person-centered messages and emotional validation via IW. These results demonstrate the importance of developing social AI chatbots that employ messages high in person-centeredness, as these messages are most important for addressing mental health concerns.