The global mental health crisis underscores the need for accessible, effective interventions. Chatbots based on generative artificial intelligence (AI), like ChatGPT, are emerging as novel solutions, but research on real-life usage is limited. We interviewed nineteen individuals about their experiences using generative AI chatbots for mental health. Participants reported high engagement and positive impacts, including better relationships and healing from trauma and loss. We developed four themes: (1) a sense of 'emotional sanctuary', (2) 'insightful guidance', particularly about relationships, (3) the 'joy of connection', and (4) comparisons between the 'AI therapist' and human therapy. Some themes echoed prior research on rule-based chatbots, while others seemed novel to generative AI. Participants emphasised the need for better safety guardrails, human-like memory and the ability to lead the therapeutic process. Generative AI chatbots may offer mental health support that feels meaningful to users, but further research is needed on safety and effectiveness.
Mental ill-health is a major and growing cause of suffering worldwide, with an estimated 970 million people living with mental disorders in 2019 (a 48% increase from 1990), and with the likelihood of developing some mental disorder by age 75 estimated to be around 50%; a picture that looks more serious still when subclinical mental disorders are included. Access to care remains limited, with for example only 23% of individuals suffering from depression receiving adequate treatment in high-income countries, while in low- and middle-income countries, the figure drops to a mere 3%.
Digital mental health interventions (DMHIs) have emerged over the last decade as a promising potential response to the treatment gap, leveraging technology to deliver low-cost, effective, always-available and anonymous (and thus low-stigma) mental health treatment at scale. Typically delivered through mobile apps and websites, DMHIs encompass a range of tools including psychoeducation, mood tracking, mindfulness, journalling, peer support and digital cognitive behavioural therapy (CBT) programs. However, the evidence for the effectiveness of DMHIs has been limited, with a meta-analysis of randomised controlled trials (RCTs) finding only small effect sizes, potential publication bias, and a lack of active controls. Moreover, user engagement remains a persistent challenge, with mixed user reviews, and studies indicating that 30 days after installation the proportion of users still active may be as low as 3%.
Rule-based AI chatbots show promise to address some of these limitations, by simulating human conversation using predefined scripts and algorithms such as decision trees, to deliver the benefits of DMHIs in a more dynamic and interactive way. For example, two popular chatbots, Woebot and Wysa, have been shown to improve users' depression symptoms, and build therapeutic alliances that appear comparable to those formed with human therapists. Rule-based chatbot apps have more promising user engagement, with positive app store ratings and qualitative studies finding that users appreciate the human-like interaction and social support. But despite these promising signs, rule-based AI chatbots still fall short in realising the full potential of DMHIs. Meta-analyses indicate that the therapeutic effects are small and not sustained over time, and users report frustration with responses that feel empty, generic, nonsensical, repetitive and constrained.
Recent developments in generative AI technologies, such as large language models (LLMs), present new possibilities. Unlike rule-based AI chatbots, generative AI chatbots like OpenAI's ChatGPT, Google's Gemini, and Inflection's Pi are trained on vast amounts of data, enabling them to understand and generate language with remarkable proficiency. These models are increasingly achieving or surpassing human performance benchmarks in various domains, including medical diagnostic dialogue, persuasive communication, theory of mind, making people feel heard, responding to relationship issues and helping people reframe negative situations to reduce negative emotions. Furthermore, user engagement has been impressive, with ChatGPT's user base growing to 100 million weekly active users within a year of launch and an estimated half of the US population having used generative AI.
Generative AI's capabilities represent a significant opportunity for digital mental health, with media reports of increasing consumer usage, one meta-analysis finding generative AI chatbots more effective than rule-based ones at reducing psychological distress, and a pilot study showing promising results from ChatGPT usage in psychiatric inpatient care. However, this new technology also brings new challenges, including potential risks of harm and questions of liability; trustworthiness issues such as the tendency to output incorrect or fabricated content (to "hallucinate"), lack of predictability or interpretability, and inherent biases in training data; and the need to demonstrate clinical effectiveness.
There is an acknowledged lack of research in this area. Given the novelty of generative AI and the nascent state of the field, qualitative research is an important starting point to generate rich foundational insights into individuals' subjective experiences, which can be overlooked in quantitative studies. Qualitative studies published so far include thematic analyses of user forum comments on both generative AI and rule-based DMHIs, student survey responses on companion-focused generative AI chatbots, and semi-structured interviews with hospital outpatients who were asked to try ChatGPT for mental health support. To our knowledge, no study so far has employed semi-structured interviews and reflexive thematic analysis to explore the research question of how people currently experience using generative AI chatbots to work on their mental health and wellbeing, in unprompted, unguided real-world settings. This study aims to fill that gap, with a view to providing insights for researchers, platform developers and clinicians into the implications of applying this new technology to mental health care.
