AI therapy apps: what they are, what the law says, what to check

There is no such thing, yet, as an AI that is your therapist. When the FDA reviewed the field in November 2025 it had authorized no AI-enabled device for mental health, and several states now restrict apps that present themselves as therapy. What does exist is a growing set of apps you can talk to. Some are useful. This page explains what they are, what the law and the research say, and what to check before you trust one.

What "AI therapy app" usually means

• A chat you can talk to. A general AI model given a supportive role. You type or speak, it answers. Most apps people find by searching are this.
• A skills coach. Walks you through exercises borrowed from therapy, like noticing a thinking trap or planning a small step. Some older apps used scripted replies rather than a generative AI.
• A clinical tool. Built to be used with, or prescribed by, a clinician. These are the ones that go through FDA review. As of the FDA's November 2025 review, none of the AI-powered ones had been authorized for mental health.

What the FDA says

The FDA held an advisory committee meeting on generative-AI mental health devices on 6 November 2025. Its own briefing document says that of the AI-enabled devices it has authorized, "none of those AI-enabled devices have been authorized for mental health uses," and that it has authorized "fewer than twenty digital mental health medical devices," none of them AI. It also notes that most consumer mental health apps "are not reviewed or authorized by FDA." So when an app talks about being clinical or proven, ask what exactly was reviewed, and by whom.

What state law now says

These laws are new and still changing, and this is not legal advice. As of October 2026:

• Illinois (2025): nobody may "provide, advertise, or otherwise offer therapy or psychotherapy services, including through the use of Internet-based artificial intelligence," unless a licensed professional provides it. Self-help resources that do not claim to offer therapy are exempt.
• Nevada (July 2025): an AI provider may not claim, "explicitly or implicitly," that its system provides professional mental or behavioral health care, or present it as a therapist, counselor or doctor.
• Utah (May 2025): regulates "mental health chatbots" rather than banning them. They must say they are AI and not human, and may not sell or share your identifiable health information.
• New York (November 2025): AI companion apps must say they are not human and must detect signs of suicidal thinking or self-harm and point people to crisis services.
• More states followed in 2026. Maine, for one, now bars offering therapy, "including through the use of Internet-based artificial intelligence," unless a licensed professional provides it.

The direction is consistent: an AI may help you talk things through, but it may not pretend to be your therapist.

What the research says so far

• The best-known trial. In a 2025 randomized trial published in NEJM AI, Dartmouth researchers gave a therapy chatbot they built to 106 people with depression, anxiety, or eating-disorder risk for four weeks, with 104 similar people as a control group. Among the people using it, depression symptoms fell 51% on average and anxiety symptoms 31%. The lead author still cautioned that "no generative AI agent is ready to operate fully autonomously in mental health."
• Across many trials, a 2026 meta-analysis of 39 randomized trials in npj Digital Medicine found small benefits for depression and anxiety compared with control groups. Most of those trials tested older, scripted chatbots; only 8 tested generative AI like today's chat apps.
• The cautions are real too. The American Psychological Association warned in November 2025 that most such tools "are not designed or intended to provide clinical feedback or treatment," "may lack scientific validation and oversight," and often lack adequate safety protocols. It suggests they can support ongoing care with a qualified provider, not replace it.

What to check before you trust one

• Does it say plainly that it is an AI? Not once in the small print, but in a way you will actually notice.
• Does it claim to be a therapist, licensed, or clinically approved? Ask what that claim rests on. "AI therapist" as a description is not a license.
• What happens if you say you are not safe? A good tool points you to real crisis help. None of them can send help on their own.
• What happens to what you tell it? Look for who can read conversations, whether your words are used for ads, and how to delete everything.
• Does it push you to keep talking, or to pay, at the wrong moment? Pressure in the middle of a hard conversation is a bad sign.
• Tell your own clinician if you have one and you use an app between sessions. That is the APA's advice, and it is good advice.

Where FeelBetterBot fits, honestly

• It is a free AI to talk to, with no sign-up, that remembers what matters to you between visits. It is not therapy, and it will not tell you it is.
• It is not FDA-reviewed and does not diagnose or treat anything.
• If you mention thoughts of suicide or self-harm, it points you to 988 while staying with what you said. It cannot call anyone for you.
• No ads, and by standing rule nobody on our team reads your conversations. The details are on our privacy page.

If what you need is an actual therapist, our talk to a real person page covers free and low-cost ways to reach one. For what chatbots in general can and cannot do, see mental health chatbots, honestly.

Sources, checked 23 October 2026: FDA, Digital Health Advisory Committee executive summary (Nov 2025) · Illinois Public Act 104-0054 · Nevada AB 406 · Utah HB 452 · New York Governor's office on AI companion safeguards · Maine 10 M.R.S. §1500-EE · Dartmouth, Therabot trial · npj Digital Medicine meta-analysis (2026) · APA health advisory on AI chatbots.