Abstract
You are seeing Robert, a 33-year-old man with depression and anxiety, who has some passive suicidal ideation, but no intention or plan. Working with you, he's started an SSRI trial and feels better, though still experiences anxiety symptoms. He's interested in psychotherapy, particularly in CBT. He is on a waitlist at his local hospital and can’t afford private therapy. He's used chatbots to find information on depression; the experience was very positive. He asks: should I try a chatbot for psychotherapy?
Patients are increasingly wondering about the role of chatbots in mental healthcare. Chatbots aren’t new but interest has grown significantly since the release of ChatGPT 3.5. In a cross-sectional study of people aged 12 to 21 years, 13% of them used generative AI for mental health advice. Perspective: that's more than 5 million U.S. youth. 1
It's easy to see the attractiveness of chatbots for psychotherapy. ChatGPT, Gemini, and other chatbots are accessible to anyone with a computer or a smartphone; as well, there are mental health-specific products. Many offer a low-frills version at no financial cost to the user. And chatbots seem very knowledgeable, offering advice for everything from cooking to travel. So why then not ask for guidance on, say, thought logs and negative thoughts?
The need is great. In Canada, many lack access to psychotherapy. In a 2016 review, Patten et al found that only about 1 in 4 people with depression received counselling or psychotherapy. 2 Chatbots thus have great potential. After experimenting with different chatbots, Dr Allen Francis concluded: “Artificial intelligence is an existential threat to our profession. Already a very tough competitor, it will become ever more imposing with increasing technical power, rapidly expanding clinical experience and widespread public familiarity.” 3 In an RCT, those who received care from an LLM trained on CBT had significant reductions of mood and anxiety symptoms over the wait-listed control; while we emphasize here that the design of the study was thus limiting given the control, we add: participants rated the therapeutic alliance as comparable to that of human therapists. 4 In a recent qualitative analysis, one theme that emerged among participants using chatbots for mental health was the “joy of connection.” 5
AI isn’t a panacea: not everyone has a computer; language barriers persist. But there is the larger question: are chatbots appropriate for psychotherapy? At this point, no AI company is making psychiatric or clinical claims about its AI, which is telling. Problems are evident.
Privacy
Who would have access to patient data? How could it be shared? Not surprisingly, in 2023, the American Psychiatric Association released an advisory suggesting that clinicians not enter any patient information into any AI chatbot. 6
Safety
Though chatbots may be engaging and supportive, there are questions about their overall safety, in part because they are programmed to get along with people. Several cases have been reported in the media of people who seek advice from AI chatbots with disastrous results. 7 There have also been lawsuits alleging bad advice. Posing as an adolescent, Dr Andrew Clark discussed different scenarios with 10 AI chatbots; the results were, at best, uneven in quality. The majority agreed, as an example, that a teenager with depression should stay in her room and not communicate with anyone. 8
Bias
Different bots have different biases in evaluating normal and abnormal behaviour. Illustrating the point: when 5 bots were asked to compare the strengths and weaknesses of the leaders of their respective companies, each bot emphasised the strengths of its own company's leader and minimized their weaknesses. 3
Things are evolving quickly. LLM models will likely become safer with more time and training; ChatGPT, as an example, has gone through major updates in recent months. In terms of privacy, AI companies have recently announced HIPAA-compliant versions that may better protect health data. In some ways, this article is already out of date.
But clinical advice to our patients shouldn’t be based on future potential but today's knowledge. How then should you answer a patient who asks about chatbots?
Psychoeducation would be appropriate. After all, LLMs are capable of generating helpful and immediate mental health information—though there is evidence that mental health advice may not be as robust as human-generated answers. 9 Patients could be advised to use chatbots for information but not for therapy itself.
What about psychotherapy? There are several well-established alternatives for engaging in CBT that do not rely on conversational AI. The strongest option remains clinician-led CBT, delivered individually or in groups by a trained professional. Patients could find options outside of the public system to be cost-prohibitive. Some, though, may have coverage through an extended health benefits plan and they should explore this option. Computer-based CBT is also safe and effective because it follows set rules and cannot go “off the rails” and say dangerous things.
For people with mild to moderate symptoms, guided self-help CBT using structured workbooks or online programs with brief human support can achieve outcomes close to full therapy. One way of doing this would be for a patient to use non-chatbot digital CBT programs that rely on preprogrammed modules and worksheets. A low-tech solution: high-quality CBT self-help books which also provide evidence-based exercises such as thought records, behavioral activation, and exposure planning without the unpredictability of AI.
Returning to Robert: After some discussion, he decides not to use a chatbot. As someone who has experienced trauma, he is particularly concerned about privacy issues. Instead, he visits his local library and looks at several CBT workbooks and decides to try Mind Over Mood, which he finds helpful and relevant. He is particularly engaged with the homework. And, after reviewing its privacy policy, he opts to download an app, allowing to track his mood (with scales) and better understand his illness with psychoeducation.
Clinical Pearls
Chatbots are accessible and engaging, leading some patients to wonder about their use for psychotherapy.
There are clear problems, including with privacy, safety, and bias.
Patients are better advised to use more traditional resources, including apps and workbooks.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.