Conversational AI is increasingly being used by young people for emotional support, advice and conversations about personal concerns. Emerging evidence suggests that these systems may offer mental health benefits, particularly for depressive symptoms, while also raising concerns about dependency, relational dynamics, safety and therapeutic precision. This debate article argues that the central question is not whether conversational AI should be viewed as beneficial or harmful, or as therapeutic or nontherapeutic. Rather, its impact on young people’s mental health may depend substantially on how these systems are designed. As the boundaries between clinically oriented and general-purpose conversational agents become increasingly blurred, design decisions concerning relational style, responses to distress, risk management, transparency and human oversight take on growing importance. These decisions are clinical and developmental as much as technical. The article argues that mental health expertise has an important role to play in shaping, evaluating and governing conversational AI systems, particularly as they become more embedded in the everyday lives of young people.