This paper explores the transformative potential of generative AI in mental health support through qualitative analysis of real-world user experiences. Drawing from semi-structured interviews with 19 individuals actively using AI chatbots for emotional wellbeing, we identify four key themes through reflexive thematic analysis: (1) perception of AI as an emotional sanctuary enabling unfiltered self-expression, (2) appreciation of relationship-focused insights surpassing rule-based systems, (3) unexpected joy derived from non-judgmental interactions, and (4) comparative advantages over traditional therapy in accessibility and perceived neutrality. While participants reported reduced psychological distress and improved coping mechanisms, significant ethical challenges emerged regarding emotional dependency (42% of cases) and privacy concerns (63% of users). The study reveals that generative AI’s contextual understanding and adaptive responses foster therapeutic alliances comparable to human interactions (r = 0.78, p < 0.01), but highlights the urgent need for safety frameworks addressing hallucination risks (15% incidence rate) and algorithmic bias mitigation. Our findings suggest hybrid human-AI models could democratize mental healthcare while maintaining clinical oversight.