A review of 16 randomised trials finds limited and inconsistent improvements from AI-delivered CBT for anxiety and depression, with dropout and bias clouding the evidence.
A new review in npj Digital Medicine has found that AI-delivered cognitive behavioural therapy, or CBT, has so far produced only limited and inconsistent benefits for anxiety and depression, even as interest in therapy chatbots continues to rise. The paper reviewed 16 randomised controlled trials involving 2,875 participants and found that just one of 14 interventions for anxiety and four of 16 for depression showed significant improvement in intention-to-treat analyses, which include everyone originally assigned to a study group. The authors say the evidence suggests the field is advancing quickly, but not yet reliably enough to justify broad confidence in these tools.
The review also challenges the assumption that more sophisticated technology automatically leads to better mental health results. According to the authors, interventions with more advanced personalisation did not outperform simpler ones, and adding human support did not clearly improve outcomes either. Early signs were more encouraging for so-called third-wave therapies, such as acceptance and commitment therapy and mindfulness-based approaches, while younger users appeared to benefit more than older adults. The review found little evidence of meaningful gains among the smaller number of studies involving older people.
Quality remains a major concern. Every trial included in the review was rated as having some concerns or a high risk of bias, mostly because of heavy dropout, which can weaken the reliability of results. The authors also noted that only one of the studies they assessed used generative AI, suggesting that much of the current evidence base still reflects earlier forms of chatbot-style therapy rather than the newest systems now attracting attention.
The findings sit alongside a broader but mixed research picture. Other recent systematic reviews and meta-analyses, including papers indexed by PubMed, have reported that chatbot-based interventions can reduce depressive and anxiety symptoms, especially when they are designed around CBT principles and used in adults with more severe symptoms. But those reviews also stress the need for better trials, stronger engagement and improved design. Taken together, the literature suggests promise, but also a clear gap between enthusiasm for AI mental health tools and the level of evidence needed to show they work consistently in practice.
Disclaimer: This content is for informational purposes only and is not intended to be a substitute for professional medical judgment, advice, diagnosis, or treatment.
