ChatGPT for mental health: a small share of messages, but a large user base?

ChatGPT for mental health: a small share of messages, but a large user base?

17 September 2025

Two recent studies shed light on the use of large language models (LLMs) like ChatGPT for mental health from very different angles. At first glance, their findings look contradictory — but are they really?

Betsy Stade and colleagues surveyed a representative US sample and found that 24% of adults reported using LLMs for mental health support. These users were younger, more often male and Black, with higher rates of depression, anxiety, and PTSD. They reported difficulty accessing traditional mental health treatment – particularly due to cost and insurance coverage – and turn to LLMs because they are free, convenient, and available whenever needed. The preprint is available via OSF.

Aaron Chatterji and colleagues analyzed over 1 million ChatGPT conversations worldwide, finding that only around 2% of messages focused on relationships, reflection, or companionship. Most usage revolved around writing, tutoring, and information-seeking. From this angle, mental health looks like a niche use case. Their working paper is available via NBER.

The two studies offer complementary perspectives: Stade et al. find that nearly a quarter of people report using LLMs (70% of which ChatGPT) for mental health, while Chatterji et al. show that only around 2% of all ChatGPT messages have relational or emotional content. The level of analysis lies at the basis for theses starkly different numbers: the first study looks at users, the second at messages.

A preliminary conclusion is therefore that many people may turn to ChatGPT for mental health occasionally. This keeps the overall share of message traffic small, while the user base could still be substantial. More research is definitely needed, but this phenomenon appears larger than it might seem at first glance and we probably shouldn’t underestimate it.

Published
Categorized as insight