People with depressive symptoms are turning to ChatGPT as an informal support system — and doing so at 2 a.m.
Researchers analyzed 187,093 ChatGPT conversations from 766 participants who each completed the PHQ-8, a standard depression screening questionnaire. Participants scoring at or above the moderate-symptom threshold used ChatGPT significantly more for mental-health topics, loneliness, and interpersonal concerns. Their conversations skewed toward late-night hours and showed recurring month-level patterns. Their language leaned on first-person singular pronouns and absolutist terms — both markers associated with depressive thinking.
The finding reframes what ChatGPT actually is for a meaningful slice of its users: not a coding assistant or a search replacement, but a private, always-on listener that asks no questions about insurance. That framing matters because the chatbot's response to these users was not meaningfully different — professional redirection was no more common in high-distress conversations than in low-distress ones.
The researchers are careful to say this data should not be used for clinical screening. A language-based model trained on the conversations predicted depression symptoms only modestly, with an AUROC of 0.591 — barely better than a coin flip. What it does show is that a behavioral pattern exists and that the infrastructure for informal mental-health support has already been built, quietly, inside a product whose maker markets it as a productivity tool.