This work examines how young adults aged 18‑25 turn to the general‑purpose conversational agent ChatGPT during moments of distress. We gathered 19,930 real‑world conversations and surveyed 158 participants, then selected five representative distress dialogues for review by ten clinicians. Distressed users showed stronger emotional engagement with ChatGPT and reported more behavioral changes after the interaction compared with their peers. When confronted with acute distress, ChatGPT tended to respond quickly with overly dramatic language and a flood of action‑oriented suggestions. Clinicians praised the system’s availability and most of its phrasing, yet identified seven process failures, the most critical being a premature jump to solutions without first calming the user. From their feedback we derived a three‑stage design guideline: (1) ask about safety to rule out self‑harm or harm to others; (2) de‑escalate emotional intensity through empathetic language to restore regulation; (3) explore the user’s concerns without endorsing them, encouraging reflection.
Review