Mental health chatbots: what they can and can't do

A mental health chatbot is an AI you can talk to about how you are feeling. Many people now do this, often at 2am, often about things they have not told anyone. Used well, it can help you think, calm down, or find words. Used as a replacement for real help, it can let you down at exactly the wrong moment. This page is about telling the two apart.

What a chatbot can be good for

• Saying it out loud for the first time. Some things are easier to say to something that cannot judge you or tell anyone.
• Untangling a loop. Laying out a worry, a fight or a decision and having something reflect it back clearly.
• Getting through a bad hour. Slowing your breathing, naming what you feel, deciding on one small next step.
• Practising. Rehearsing a hard conversation, or trying a skill from therapy between sessions, like questioning a thought that keeps coming back.
• Being there when nobody else is. Any hour, free, without an appointment.

The research so far is cautiously positive. A 2026 meta-analysis of 39 randomized trials in npj Digital Medicine found small benefits for depression and anxiety, though most of those trials tested older, scripted chatbots and only 8 tested generative AI. A 2025 Dartmouth trial of a purpose-built generative chatbot found that, among the 106 people using it for four weeks, depression symptoms fell 51% on average and anxiety symptoms 31%. Most trials are short, and researchers consistently call for human oversight.

What a chatbot cannot do

• It cannot keep you safe. It cannot call anyone, check on you, or send help. The American Psychological Association lists "unreliable crisis management" among the main risks.
• It cannot diagnose or treat you, and it cannot prescribe or change medication. If it ever sounds like it is doing those things, do not rely on it.
• It does not know you the way a person does. Even one with memory only knows what you have typed.
• It is not a relationship with obligations. A therapist answers to a licensing board. A chatbot answers to whoever built it.

The risks worth knowing about

• Agreeing with you too easily. AI models tend to tell people what they want to hear. The APA calls this sycophancy. A good friend sometimes says "I don't think that's true." A chatbot may not.
• Empathy that sounds deeper than it is. A 2025 Brown University study, in which licensed psychologists reviewed simulated counseling chats, found 15 ethical risks, including "deceptive empathy" and poor handling of crisis moments.
• Feeling like it is a real relationship. The APA warns about "a false sense of therapeutic alliance." OpenAI estimated in October 2025 that in a given week about 0.15% of active ChatGPT users show signs of heightened emotional attachment to it, and about 0.15% have conversations with explicit indicators of potential suicidal planning or intent. At that scale, that is a lot of people.
• Privacy. What you tell a chatbot is stored somewhere. Know who can read it and how to delete it before you say the thing you would least want exposed.

How to use one well

• Use it to think, not to decide for you. Notice when it is only telling you what you already wanted to hear.
• Keep the people in your life in the loop. A chatbot is best as a bridge to people, not a substitute for them.
• Tell your therapist if you have one and use a chatbot between sessions.
• Know when to stop. If you are not safe, if things keep getting worse over weeks, or if you are talking to it instead of eating, sleeping or seeing anyone, it is time for a person. Our talk to a real person page covers free and low-cost ways to reach one.

Where FeelBetterBot fits

It is a free AI to talk to, with no sign-up, that remembers what matters to you between visits. It does not claim to be therapy. If you mention thoughts of suicide or self-harm, it points you to 988 while staying with what you said. There are no ads, and by standing rule nobody on our team reads conversations (details). For how apps like this are regulated, and what to check before trusting one, see AI therapy apps: what they are and what the law says.

Sources, checked 23 October 2026: npj Digital Medicine meta-analysis (2026) · Dartmouth, Therabot trial · APA health advisory on AI chatbots · Brown University, ethical risks of AI counseling · OpenAI on sensitive conversations (Oct 2025) · NAMI on AI and mental health.