ChatGPT as therapy: A qualitative and network-based thematic profiling of shared experiences, attitudes, and beliefs on Reddit

In recent years, large language models (LLMs), including ChatGPT, have exponentially grown in application. Given existing barriers to mental health services, alongside the capability of LLMs to generate therapeutic responses, LLMs have potential to serve as accessible precursors, adjuncts, or alternatives to traditional therapy. However, little is known about the opinions of persons who have used LLMs for mental health-related problems. Thus, the current work investigated the positive and negative experiences of those who have interacted with ChatGPT for their mental health using posts from relevant Reddit threads (N = 1,594). A 33-item coding scheme was applied to code the presence/absence of each item, and coded posts were modeled using an Ising network graph to explore pairwise and groupwise thematic associations of items. Results from the qualitative coding indicated that the codes representing positive sentiment, experiencing affect/emotion, and attaining personal benefit from using ChatGPT for therapy were among the most frequently coded. Moreover, the Ising network model revealed the most important associations in the network between items representing positive experiences, negative experiences, and specific pathologies. Taken together, the current study indicates that, while users endorse several benefits from using ChatGPT for their mental health, they also report significant drawbacks, including restrictions by ChatGPT that can harm them or exacerbate their symptoms. Thus, ChatGPT may be a useful tool for persons not able to receive standard care, but it may not be as beneficial for persons with more severe pathologies or mental health concerns.