The increasing use of artificial intelligence (AI)-powered companion applications is transforming how people engage with digital entities. AI companions can serve various roles that foster deep emotional connections, including friend, mentor, or romantic partner. These applications are growing in popularity, raising questions about their impact on users'. While some people report finding comfort and support in AI companionship, reports of uncomfortable interactions underscore the urgent need for comprehensive evaluation and responsible design. This project investigates the dynamics and quality of interactions between people and AI companion apps, ensuring these technologies enhance user experience. By understanding user experiences, assessing the quality of AI-driven conversations, and developing design guidelines, and promoting their safe and responsible deployment. This research will be conducted in three phases. First, large-scale social media data and user interviews will be analyzed to develop a taxonomy of user interactions, identifying motivations, challenges, and values associated with AI companion apps. Second, an innovative algorithmic audit will be conducted using role-playing AI personas, which are LLM-based simulated users representing diverse psychological and behavioral profiles, to evaluate the quality of conversations with AI companion applications. Third, participatory design workshops will engage users in co-creating guidelines for ethically aligned AI