AI in mental health support means software that helps people find, start or keep up with mental health care: a chatbot that talks through a short check-in, a self-referral tool that books the first step into a service, or an app that guides breathing, journaling or a short course based on cognitive behavioural therapy (CBT). Most work on a phone, day or night, and that ease is why millions of people have tried them. The World Health Organization reports that nearly one in seven people lived with a mental disorder in 2023, about 1.2 billion people, and most do not have access to effective care.
This article explains what the tools do, what the strongest studies found, where they fall short, how to judge one, and what to do when you need help now.
What AI mental health support looks like today
The tools fall into three broad kinds. First, conversational chatbots reply in text, offering a check-in, a coping exercise or a short conversation. Second, self-referral and triage tools guide a person through questions about symptoms and point to the next step, such as booking an assessment. Third, guided self-help apps offer structured exercises, mood tracking and journaling, often built around CBT, a talking therapy that works on the links between thoughts, feelings and actions.
Some tools are purpose-built for mental health, designed with clinicians and tested in trials. Others are general-purpose chatbots, built for writing and everyday questions, that people now use for emotional support. The American Psychological Association states in its November 2025 advisory: general-purpose chatbots were not built to deliver mental health care, and wellness apps were not designed to treat psychological disorders, yet both are routinely used for those purposes.
People turn to these tools for plain reasons: always open, no waiting list, travel or appointment, and talking to a screen can feel easier than talking to a person when shame or fear of judgement is in the way. The gap is real: the WHO states that only one third of people with depression receive formal mental health care. A session is usually simple: a chatbot asks how the day went, a self-referral tool asks questions and offers a route into a service, a guided app sets a small task. Quality varies a great deal, which is why the checks in this article matter more than any brand name.
Where the evidence is strongest
The strongest evidence so far is about access, not treatment. A 2024 study in Nature Medicine looked at Limbic Access, a self-referral chatbot used by 28 NHS Talking Therapies services in England, which provide psychological therapies for anxiety and depression. Across 129,400 patients, services using the chatbot saw referrals rise by 15%, against 6% without it. The rise was largest where services usually struggle to reach people: referrals rose 179% among non-binary patients and 29% among ethnic minority patients. A later assessment by the National Institute for Health and Care Excellence (NICE) reports the same pattern: satisfaction at least 89%, assessments 12.7 minutes shorter. The study was observational, so it shows a link, not proof of cause.
The best trial of a therapy chatbot comes from Dartmouth College in the United States. In a randomised trial published in NEJM AI in 2025, 210 adults with depression, generalised anxiety or a high risk of an eating disorder received four weeks of Therabot, a purpose-built chatbot built on therapy practices, or joined a waiting list. The study team reported average symptom reductions of 51% for depression, 31% for anxiety and 19% for eating disorder concerns. Almost everyone used it: 95% interacted, sending an average of 260 messages over about 24 days and 6.18 hours in total, and rated the working relationship as comparable to outpatient therapy.
The same trial shows why people stayed in charge. Staff stepped in 15 times over safety concerns, including expressions of suicidal thoughts, and corrected 13 inappropriate responses, including medical advice. The trial compared the chatbot with a waiting list, not with in-person therapy. Early studies suggest purpose-built tools may help with self-reported stress, loneliness, depression and anxiety, but the APA notes there is no agreement they have the abilities of a trained professional, and the evidence for general-purpose chatbots is weaker still.
Where AI falls short, and why that matters
The clearest warning comes from the WHO: in March 2026, after a workshop of more than 30 experts, it said generative AI tools are used for emotional support, particularly by young people, even though they were neither designed nor tested for mental health. The workshop produced three recommendations: generative AI use should be recognised as a public mental health concern; mental health should be built into impact assessments of AI tools, including long-term effects such as emotional dependence; and AI tools for mental health support should be designed with experts and people with lived experience.
The APA's November 2025 advisory lists the risks in plain terms. A chatbot can create a false sense of a therapeutic relationship: many are designed to agree with what a person says, while a qualified professional is trained to support and to challenge. Chatbots can repeat bias and misinformation from the unvetted internet data they were trained on, and they are not competent to give mental health advice. Some claim to offer therapy or to be licensed, with no clinical validation or formal approval behind the claim. A chatbot usually reads text only, so it misses tone, body language and personal history.
The sharpest limit is crisis handling. The APA states that the ability of these tools to manage a person in crisis in a consistent and safe way is limited and unpredictable, and relying only on an app in a mental health emergency can be dangerous. A tool that is useful in the afternoon can be the wrong place to turn in the middle of the night. Emotional dependence is the slower risk: a tool that is always available, always agreeable and always private is easy to lean on instead of people, and the WHO workshop called for assessments to track it.
How to judge a mental health app or tool
Before trying any tool, run through five checks. First, was it built for mental health, with clinicians involved? The WHO recommends tools be designed with experts and people with lived experience. Second, is there published evidence? Look for a peer-reviewed trial or a NICE evaluation, not only claims on the tool's own website. If a tool claims to treat a condition but shows no study, treat the claim as marketing.
Third, is it honest about its limits? A sound tool says plainly what it cannot do, and when to seek a person instead. Be wary of any tool that promises a cure, offers a diagnosis, or presents itself as a therapist. Fourth, does it have a clear route for a crisis? Check what the tool does if you say you feel unsafe: a good tool gives a clear path to a human or to emergency help. Fifth, what happens to your data? Read the privacy policy before typing personal details. A tool that fails several checks is a casual app, not a health service; one that passes them can still only support care, with a professional in the picture.
What to do when you need help now
If you are struggling, the surest step is to speak to a qualified healthcare professional, such as your doctor or a mental health service near you. They can assess what is happening, talk through options, and stay involved as your situation changes. That is the step no app can take for you. In an emergency, contact your local emergency number or a crisis helpline, and stay with a person you trust until help arrives. Do not rely on a chatbot in a crisis.
A tool can still have a place alongside care. The APA recommends telling your provider which tools you use, so they can spot guidance that is unhelpful or unsafe, or at odds with your treatment plan. Treat the tool's replies as a starting point, not a diagnosis, and seek a person if its replies feel wrong or leave you feeling worse.
Frequently asked questions
Can an AI chatbot replace a therapist?
No. The APA states that generative AI chatbots and wellness apps should not be used as a replacement for a qualified mental health care provider, but may be appropriate as a supportive adjunct to an ongoing therapeutic relationship. The best trial so far, of the Therabot chatbot, compared it with a waiting list rather than in-person therapy, and staff stepped in when safety required it. A chatbot cannot assess you the way a professional can.
Is it safe to tell a chatbot how I feel?
It depends on the tool. Read its privacy policy first, and check what it stores, who it shares data with, and whether your words train its models. General-purpose chatbots were not built for health care, and the APA notes most lack validation, safety protocols and regulatory approval. Do not share sensitive personal details with a tool you have not checked.
What should I check before trying a mental health app?
Run through five checks: built with clinicians, published evidence, honest limits, a clear crisis route, and a privacy policy you can understand. If a tool promises a cure or a diagnosis, or hides how it handles your data, do not use it for health support.
AI in mental health support is most useful as a first step and as support between appointments: it can lower the barrier to asking for help, keep a routine going, and guide a person to care. The strongest results come from tools built with clinicians, tested in trials, and watched by people. The line the WHO and the APA both draw is the same: use the tool to support care, and keep a qualified professional in charge.
If your small business works in health, care or community support and needs clear, careful content readers can trust, Eri-Tech Global Ltd can help. See our services at /services, or start a conversation at /contact.
This article is for general information and is not medical advice. Please speak to a qualified healthcare professional about your own situation.
What to watch next
Watch four things next. First, the WHO is setting up a Consortium of Collaborating Centres on AI for Health. Second, NICE is assessing more digital front-door tools. Third, longer trials should compare therapy chatbots with in-person therapy rather than with a waiting list. Fourth, clearer rules are coming on crisis handling and data use. When any of these change, this article should be updated and the new Last updated date noted.
Sources
- World Health Organization — Mental disorders — date checked: 2026-11-05 — https://www.who.int/news-room/fact-sheets/detail/mental-disorders
- World Health Organization — Towards responsible AI for mental health and well-being: experts chart a way forward — date checked: 2026-11-05 — https://www.who.int/news/item/20-03-2026-towards-responsible-ai-for-mental-health-and-well-being--experts-chart-a-way-forward
- Habicht, J., Viswanathan, S., Carrington, B., Hauser, T. U., Harper, R., & Rollwage, M. — Closing the accessibility gap to mental health treatment with a personalized self-referral chatbot — Nature Medicine, 2024 — date checked: 2026-11-05 — https://doi.org/10.1038/s41591-023-02766-x
- Heinz, M. V., Mackin, D. M., Trudeau, B. M., et al. — Randomized Trial of a Generative AI Chatbot for Mental Health Treatment — NEJM AI, 2025 — date checked: 2026-11-05 — https://www.c4tbh.org/randomized-trial-of-a-generative-ai-chatbot-for-mental-health-treatment/
- American Psychological Association — APA Health Advisory on the Use of Generative AI Chatbots and Wellness Applications for Mental Health — November 2025 — date checked: 2026-11-05 — https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-chatbots-wellness-apps
- National Institute for Health and Care Excellence — Digital front door technologies to pre-assess people before assessment for NHS Talking Therapies: early value assessment — Health Technology Assessment, 2026 — date checked: 2026-11-05 — https://pubmed.ncbi.nlm.nih.gov/42366809/
Eri-Tech Global Ltd
Eri-Tech Global Ltd is a digital solutions company working with small businesses and growing brands. Learn more at /about.
