AI Chatbots for Depression: What Does the Evidence Show?

How much can conversational agents improve mental health symptoms?

The systematic review, published in npj Digital Medicine on May 29, 2026, found standardized mean differences of −0.27 for depression, −0.20 for anxiety and −0.26 for stress. These measure differences between intervention and comparison groups on a common scale; they are not percentage reductions or recovery rates. [Read the meta-analysis](https://www.nature.com/articles/s41746-026-02820-1).

Evidence certainty was moderate for depression and stress but low for anxiety, partly because of possible publication bias. Long-term benefit remains uncertain. Both scripted and AI-based systems were included, so the findings should not be generalized to every contemporary generative chatbot. [Study findings and limitations](https://www.nature.com/articles/s41746-026-02820-1).

How might a mental health chatbot help someone feel better?

Cognitive behavioral therapy, or CBT, helps people recognize unhelpful interpretations, examine the evidence behind them and change behaviors that reinforce distress. For example, someone who interprets a setback as proof of personal failure might practice developing a more balanced explanation. The National Institute of Mental Health describes these techniques as components of established psychotherapy. Delivering an exercise through software changes its format; the underlying therapeutic skill still matters. [NIMH guidance on psychotherapy](https://www.nimh.nih.gov/health/topics/psychotherapies).

Digital tools can also support practice between appointments, offer coping exercises and help users record symptoms. Their practical appeal includes access outside clinic hours and reduced travel demands. NIMH emphasizes that engagement matters: an intervention has limited opportunity to help if people stop using it. A reasonable interpretation is that useful therapeutic content and sustained participation deserve attention alongside how naturally a chatbot converses. [NIMH overview of mental health technology](https://www.nimh.nih.gov/health/topics/technology-and-the-future-of-mental-health-treatment).

What should patients look for before using a mental health chatbot?

A claim that an app uses CBT does not demonstrate that the complete product has been tested successfully. NIMH recommends examining supporting research, considering the developer's qualifications and checking what help is available if symptoms worsen. Discussing an app with a clinician can help establish whether it fits an existing care plan. The relevant question is whether the particular intervention helps people with comparable needs. [NIMH advice on evaluating mental health apps](https://www.nimh.nih.gov/health/topics/technology-and-the-future-of-mental-health-treatment).

Generative systems introduce additional concerns because fluent responses can contain incorrect or biased information. The World Health Organization also identifies excessive trust in automated outputs and risks to sensitive health data. Applied to mental health, this means conversational warmth should not be treated as proof of clinical accuracy or confidentiality. Product evaluation should address harmful responses and data protection as well as symptom improvement. [WHO guidance on generative AI in health](https://www.who.int/news/item/18-01-2024-who-releases-ai-ethics-and-governance-guidance-for-large-multi-modal-models).