AI Chatbots for Depression: How Much Do They Help?

What did the research find about mental health chatbots?

The review, published in npj Digital Medicine in May, examined both AI and rule-based systems. Pooled standardized effects were −0.27 for depression, −0.20 for anxiety and −0.26 for stress, with negative values favoring the interventions. Different subsets of trials contributed to each outcome. [Read the meta-analysis](https://www.nature.com/articles/s41746-026-02820-1).

These figures describe differences on symptom scales, not percentages of people cured. A depression effect of −0.27 does not mean symptoms improved by 27%. The findings support potential benefit, but do not establish that every chatbot works or that improvements persist over the long term.

Can an AI chatbot provide the same care as a therapist?

There is a clinical rationale for delivering structured psychological exercises digitally. Cognitive behavioral therapy helps people recognize unhelpful thinking and change behaviors that maintain distress. Other therapeutic elements include problem-solving, relaxation and tracking emotions. A digital conversation could help someone practice these skills, but whether a particular product delivers them effectively requires evidence about that product. The National Institute of Mental Health describes these approaches in its [psychotherapy guidance](https://www.nimh.nih.gov/health/topics/psychotherapies).

Professional care also includes assessment and treatment decisions. Depression-like symptoms can arise from medical conditions or medications, and clinicians consider severity, history and daily functioning when recommending treatment. NIMH identifies psychotherapy, medication or their combination as usual depression treatments, and notes that digital tools may supplement therapy. The practical implication is that an app's availability or reassuring tone cannot establish its ability to make those clinical judgments. [NIMH depression guidance](https://www.nimh.nih.gov/health/publications/depression).

How can patients assess a mental health chatbot safely?

The World Health Organization warns that language models can produce convincing but incorrect health information and may inadequately protect sensitive information. Its guidance calls for rigorous evaluation and evidence of benefit before widespread healthcare use. For patients, a practical implication is to ask whether research tested the actual product and intended use, rather than relying on broad claims about AI or therapy. [WHO guidance on AI for health](https://www.who.int/news/item/16-05-2023-who-calls-for-safe-and-ethical-ai-for-health).

Before sharing intimate details, check how conversations are stored, whether they are shared or used for training, and how deletion works. Also check whether human support is available when symptoms worsen. These are practical checks prompted by WHO's concerns about privacy, oversight and safety; a friendly conversation does not answer them. Persistent symptoms or difficulty managing daily life warrant assessment by a healthcare professional. Medication changes should also be discussed with the prescribing clinician. [NIMH advice on depression care](https://www.nimh.nih.gov/health/publications/depression).