CCBT and AI: A Revolution in Mental Health

Neurotech · Andreas Grivas, MSc Clinical Psychologist · August 15, 2023

CCBT and AI: A Revolution in Mental Health

Computerized Cognitive Behavioral Therapy (CCBT), often called internet-delivered CBT (iCBT), is not new. Structured online programmes that teach people to notice and challenge unhelpful thoughts, re-engage with rewarding activities and gradually face what they avoid have existed for at least 20 years and have been tested for many psychiatric and somatic conditions (Andersson et al., 2019). The evidence is strongest for anxiety and depression: therapist-guided iCBT is more effective than a waiting list and tends to be as effective as face-to-face CBT (Andersson et al., 2019). What is new is the arrival of artificial intelligence (AI), and especially large language models, which promise to turn static modules into adaptive, conversational support. Whether this amounts to a “revolution” depends on what the evidence actually shows.

It helps to understand what classic CCBT already does well and where it falls short. An individual patient data network meta-analysis pooled results from 39 trials including 9,751 people with depression. Both guided and unguided iCBT outperformed control conditions in the short and long term. Guided iCBT, in which a clinician provides brief support, produced slightly better results at the end of treatment (a mean difference of 0.8 points on the PHQ-9 depression questionnaire), but there was no evidence of a difference at 6 or 12 months. Crucially, the benefit of guidance depended on severity: for people with mild or subthreshold symptoms the difference between guided and unguided iCBT was small, whereas people with moderate to severe depression did better with guidance (Karyotaki et al., 2021). In other words, the same programme does not suit everyone equally, and matching the level of support to the person matters.

This is exactly the gap AI is expected to fill. Earlier mental health chatbots were largely rule-based, following scripted decision trees. Newer systems use generative AI to produce free-text responses that can adapt to what a person writes, remember context within a conversation and offer CBT techniques at the moment they are needed. Beyond conversation, smartphones and AI create opportunities for “digital phenotyping” (inferring changes in mood or behaviour from patterns of device use) and for remote intervention, alongside other technologies such as virtual reality. A major review in World Psychiatry concluded that these technologies are already changing mental health care, each with an early but promising evidence base, while stressing that substantial implementation challenges remain (Torous et al., 2021).

What does the research show so far? A systematic review and meta-analysis of AI-based conversational agents identified 35 studies, 15 of which were randomized controlled trials suitable for meta-analysis. AI chatbots significantly reduced symptoms of depression (g = 0.64) and psychological distress (g = 0.70), with larger effects for multimodal and generative AI-based agents, for agents integrated into mobile or messaging apps, and for clinical, subclinical and older populations. However, they did not significantly improve overall psychological well-being, and the confidence intervals were wide (for depression, 0.17 to 1.12), which means the true effect could be anywhere from small to large (Li et al., 2023). The authors also noted that users’ experience depended largely on the quality of the human–AI relationship, engagement with the content and effective communication.

The most widely discussed recent study is the first randomized trial of a fully generative AI therapy chatbot. Therabot, a chatbot fine-tuned by clinical experts, was tested in 210 adults with clinically significant symptoms of major depressive disorder, generalized anxiety disorder or high risk for eating disorders. Compared with a waitlist control, participants who used Therabot for four weeks showed greater reductions in symptoms of depression (d = 0.85–0.90), anxiety (d = 0.79–0.84) and eating disorder risk (d = 0.63–0.82) at the end of treatment and at eight-week follow-up. Participants used the app for more than six hours on average and rated their therapeutic alliance as comparable to that reported with human therapists (Heinz et al., 2025). These results are encouraging, but the authors themselves emphasize that larger clinical samples are needed. The comparison was a waitlist rather than an active treatment, the follow-up was short and the sample was modest in size.

AI also introduces risks that classic CCBT did not. Psychotherapy is an unusually high-stakes field for AI, because responsible, evidence-based treatment requires nuanced clinical expertise (Stade et al., 2024). Generative models can produce responses that sound confident but are inaccurate or clinically inappropriate, may fail to recognise signs of risk such as suicidal thinking, and can reflect biases in their training data. Stade and colleagues (2024) argue that clinical AI should be developed around clinical science, through interdisciplinary collaboration, with careful attention to assessment, risk detection, transparency and bias. Data privacy is another major concern, since conversations about mental health contain highly sensitive information. Even with well-established iCBT, a significant minority of users experience negative effects (Andersson et al., 2019), so careful monitoring remains essential.

For these reasons, AI-based CCBT is best suited to people with mild to moderate symptoms of anxiety, depression or stress who want structured self-help or support between therapy sessions. It is not a suitable stand-alone treatment for people at risk of suicide or self-harm, those experiencing psychosis or mania, people with severe eating disorders or substance dependence, or anyone whose safety is at risk. If you are in crisis, have thoughts of ending your life or feel unable to keep yourself safe, contact your local emergency number immediately rather than turning to an app.

It is also wise to seek an assessment from a mental health professional if your symptoms are not improving after several weeks of digital self-help, if they are getting worse, or if they significantly affect your work, relationships or daily functioning. A clinician can make a diagnosis, consider medication where appropriate and offer more intensive therapy. In a well-organised stepped-care system, AI tools can serve as a first step or a companion to treatment; they complement, but do not replace, professional care.

AI is changing how CBT can be delivered, making it more conversational, more personalised and available at any hour. Early trials, including the first randomized trial of a generative AI therapy chatbot, show real promise for reducing depressive and anxiety symptoms. Calling it a revolution, however, would be premature: the evidence is still young, the risks are real, and the safest path forward combines the reach of AI with clinical oversight and rigorous research.

References

  • Andersson, G., Titov, N., Dear, B. F., Rozental, A., & Carlbring, P. (2019). Internet-delivered psychological treatments: From innovation to implementation. World Psychiatry, 18(1), 20–28. https://doi.org/10.1002/wps.20610
  • Heinz, M. V., Mackin, D. M., Trudeau, B. M., Bhattacharya, S., Wang, Y., Banta, H. A., Jewett, A. D., Salzhauer, A. J., Griffin, T. Z., & Jacobson, N. C. (2025). Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI, 2(4), AIoa2400802. https://doi.org/10.1056/AIoa2400802
  • Karyotaki, E., Efthimiou, O., Miguel, C., Bermpohl, F. M. G., Furukawa, T. A., Cuijpers, P., & Individual Patient Data Meta-Analyses for Depression (IPDMA-DE) Collaboration. (2021). Internet-based cognitive behavioral therapy for depression: A systematic review and individual patient data network meta-analysis. JAMA Psychiatry, 78(4), 361–371. https://doi.org/10.1001/jamapsychiatry.2020.4364
  • Li, H., Zhang, R., Lee, Y.-C., Kraut, R. E., & Mohr, D. C. (2023). Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. npj Digital Medicine, 6, 236. https://doi.org/10.1038/s41746-023-00979-5
  • Stade, E. C., Stirman, S. W., Ungar, L. H., Boland, C. L., Schwartz, H. A., Yaden, D. B., Sedoc, J., DeRubeis, R. J., Willer, R., & Eichstaedt, J. C. (2024). Large language models could change the future of behavioral healthcare: A proposal for responsible development and evaluation. npj Mental Health Research, 3, 12. https://doi.org/10.1038/s44184-024-00056-z
  • Torous, J., Bucci, S., Bell, I. H., Kessing, L. V., Faurholt-Jepsen, M., Whelan, P., Carvalho, A. F., Keshavan, M., Linardon, J., & Firth, J. (2021). The growing field of digital psychiatry: Current evidence and the future of apps, social media, chatbots, and virtual reality. World Psychiatry, 20(3), 318–335. https://doi.org/10.1002/wps.20883