I used to think of therapy chatbots as a gimmick. Then I read about a 32-year-old mom in Massachusetts who uses Claude to organize her thoughts before therapy appointments. A psychiatrist in Boston who uses ChatGPT to decompress after difficult patient sessions. Neither of them treats AI as a replacement for human care. But both of them are doing something the mental health system simply can’t offer right now; getting support the moment they need it.
That tension between what conversational AI can genuinely do and what people desperately want it to do is where this entire conversation gets interesting.
Key Takeaways
- 16% of adults used AI tools for mental health in the past year, per a KFF poll.
- AI companion apps grew 700% between 2022 and mid-2025, from 16 apps to 128.
- Therabot’s clinical trial showed a 51% average reduction in depression symptoms over an 8-week trial period.
- Two-thirds of teenagers now interact with chatbots, per Pew Research.
- Maine, Illinois, and Nevada have banned AI from providing standalone mental health therapy.
- AI mental health tools work best as a bridge to professional care, not a substitute for it.
What Is Conversational AI and Why Are People Using It for Mental Health?
Conversational AI refers to the software that understands and responds to natural language, the kind you’d use with ChatGPT or some other mental health chatbot. Unlike older rule-based bots with rigid menus, modern conversational AI chatbots use large language models to generate responses that feel contextual and human.
The reason people are turning to AI mental health tools isn’t that it’s complicated, but the system isn’t keeping up. In the US, 1 in 8 adolescents doesn’t get mental health treatment despite needing it. Wait times for therapists stretch weeks or months. And even when people get access, sessions are 50 minutes a week, which leaves a lot of time in between where people are on their own.
A KFF poll found that 16% of US adults used AI tools for mental health in the past year. That number skewed higher for younger adults. And nearly half of people with mental health conditions who use AI tools do so specifically for psychological support.
Why people prefer AI over traditional therapy
The appeal is not hard to understand when you look at what AI offers versus what traditional therapy can realistically provide.
| Factor | AI Chatbot | Human Therapy |
| Cost | Free or low-cost | $100–$300/session without insurance |
| Availability | 24/7, instant | Weekly sessions, limited hours |
| Stigma | Zero: no one has to know | Still carries social stigma for many |
| Wait time | None | Weeks to months |
| Crisis support | Available anytime | Depends on therapist and availability |
As per Statista, 50% of users cite 24/7 convenience as their top reason for using AI mental health tools, and 41% point to privacy.
How Big Is the AI Mental Health Market Right Now?
The numbers here are harder to ignore than the skeptics would like.
- 700% growth in AI companion apps between 2022 and mid-2025, growing from 16 to 128 apps, according to TechCrunch data cited by the APA.
- Character.AI has 20 million monthly users, more than half of them under 24.
- Slingshot AI raised $93 million in seed funding to build its clinically focused AI therapy tool, Ash.
Who is using AI therapy tools the most?
The demographic picture is pretty clear, younger people are driving adoption, and a significant portion of them are using AI for mental health specifically.
- 1 in 5 adults aged 18–29 use LLMs for mental health support.
- 2/3rd of teenagers interact with chatbots, with 28% doing so daily.
- People with anxiety and depression show particularly high uptake.
Here’s a pattern that comes out; people who are already struggling are the ones most likely to use AI mental health tools. That creates both the biggest opportunity and the biggest risk in this space.
How Does a Mental Health Chatbot Actually Work?
Most general-purpose LLM-based chatbots (ChatGPT, Claude, Gemini) generate responses based on patterns learned from massive text datasets. They’re not trained on clinical protocols or validated therapy frameworks. They respond fluidly, but without a defined therapeutic structure.
Clinically trained mental health chatbots are different. Tools like Therabot, developed at Dartmouth’s AI and Mental Health Lab, are built with evidence-based therapy techniques baked in and go through clinical validation before deployment. Usage of LLM-based chatbots surged to 45% of new mental health AI studies in 2024, up from far less in prior years.
General AI chatbots vs. clinically trained mental health chatbots
| Feature | ChatGPT / Claude / Gemini | Therabot / Ash |
| Training data | General internet text | Clinical protocols, therapy frameworks |
| Clinical validation | None | Yes (Therabot: published RCT in NEJM AI) |
| FDA status | Not applicable | Under review for some clinical tools |
| Use case | General emotional support, reflection | Structured mental health intervention |
| Safety protocols | Basic flagging | Clinical oversight and escalation paths |
The distinction matters a lot, especially when someone’s in crisis.
The Real Benefits of Conversational AI Chatbots for Emotional Support
There’s a reflex in the mental health community to dismiss AI therapy as inherently dangerous. That’s not what the data says, at least not entirely.
Dartmouth’s clinical trial of Therabot, the first randomized controlled trial of a generative AI therapy chatbot, enrolled 210 adults with major depressive disorder, generalized anxiety disorder, and eating disorder risk. Participants had four weeks of access to Therabot, and outcomes were measured at the 8-week mark:
- People with depression saw a 51% average reduction in symptoms.
- Users reported a therapeutic alliance with Therabot comparable to that of human therapists.
Those aren’t small numbers. Nicholas Jacobson, the study’s senior author and an associate professor at Dartmouth, noted that the results were comparable to what is reported for traditional outpatient therapy. A RAND survey found that 93% of people who used AI mental health tools found the advice helpful.
The benefits that show up most consistently across AI counseling research aren’t that surprising once you think about them.
Around-the-clock access for people in crisis
Anxiety at 3 a.m. is a real issue. It’s the time when you can’t call a therapist, texting a friend feels like a burden, and the thoughts loop with nothing to interrupt them. AI mental health tools can provide a response. Not a diagnosis, not crisis intervention, but a structured, non-judgmental place to put thoughts down at exactly that moment.
Nick Jacobson’s work at Dartmouth specifically cites this use case. The mental health provider shortage means that people often wait weeks for care they need now. AI fills some of that gap, imperfectly but not inconsequentially.
AI as a bridge between therapy sessions
This is the use case I find most credible. Suzi Sanford, the marketing manager in Fitchburg, Massachusetts, doesn’t use Claude as a replacement for her therapist. She uses it to piece together “fractured” thoughts in the days between sessions, then shows up better prepared.
A JAMA Psychiatry paper urged therapists to ask their patients about AI use with curiosity rather than alarm. And therapists who ignore it are missing important clinical information.
Dr. Christine Crawford put it plainly, AI conversations give her a better sense of what’s inside my patient’s brain, a jumping-off point for better in-person connection.
The Serious Risks and Limitations of AI Therapy and AI Counseling
The optimistic framing above has real limits, and I’d be doing you a disservice if I didn’t address them directly.
The APA published an advisory warning of significant harm from using generic AI chatbots for mental health support. The Lancet Psychiatry noted that clinical effectiveness remains insufficiently established across most AI mental health tools. And only 16% of LLM-based mental health studies have actually tested for clinical efficacy. The rest are measuring engagement, satisfaction, or other proxies.
Can an AI Therapist Do More Harm Than Good?
There have been lawsuits linking AI chatbot interactions to suicides and to users whose delusional thinking was reinforced. A 2026 study found that extreme use of generative AI tools was associated with delusion-like experiences in some users.
The combination of 24/7 availability, emotionally attuned responses, and no clinical oversight creates a scenario where someone in genuine distress gets responses that feel supportive but aren’t therapeutic. And a chronically depressed person may not recognize the difference.
Data privacy and dependency concerns
Dr. Crawford and other clinicians raise two concerns.
Data privacy: When you share mental health struggles with a chatbot, that data goes somewhere. Most general-purpose platforms don’t have the same privacy obligations as licensed health providers. Sharing sensitive psychological content with an AI platform may expose data in ways users don’t fully understand.
Dependency: There’s a pattern showing up in clinical discussions. People who start using AI for emotional support sometimes find it easier to talk to a bot than to another person. WBUR’s reporting includes a user who realized, gradually, that their AI conversations had started replacing human connection rather than supplementing it.
What Therapists and Mental Health Experts Actually Think About AI Counseling
The clinical community’s position on mental health chatbots is more nuanced. It’s not uniformly against AI mental health tools. It’s against the specific version where AI operates without oversight, without clinical validation, and without transparency.
Dr. Christine Crawford, Boston psychiatrist: Uses ChatGPT herself to process difficult patient sessions. Sees value in asking patients about their AI use as a clinical window into their inner world.
Nick Jacobson, Dartmouth: He led the Therabot trial. Argues AI-powered therapy may offer clinically meaningful benefits, but stresses that clinical oversight is still in critical need.
Dr. John Torous, digital psychiatry researcher at Beth Israel Deaconess: He has been vocal about the need for standardized benchmarking of AI chatbot responses before widespread clinical adoption.
One concept gaining traction is the mindbench.ai “nutrition label,” a standardized way to evaluate AI mental health tools the same way we’d evaluate a medication: what’s in it, what’s been tested, what it’s not meant to treat.
Is AI Therapy Legal? Regulations and Policy Landscape in 2026
The legal situation is moving fast and creating a genuine patchwork.
Illinois, Nevada, and Utah became the first states to formally regulate AI mental health therapy. Illinois banned standalone AI-based therapy outright. Nevada’s AB 406 forbids AI systems from providing or claiming to provide mental or behavioral healthcare. Utah took a softer approach, requiring disclosure and data protections rather than outright prohibition.
Maine went further in April 2026 with a law that now bans AI chatbots from presenting as therapists or mental health counselors. Massachusetts, New York, and Pennsylvania have introduced similar bills, largely mirroring Illinois’s approach.
The key distinction across almost every law is the same. AI can support administrative tasks like scheduling, notes, and billing, but clinical treatment must involve a licensed human. General-purpose LLMs like ChatGPT or Claude aren’t the target of these laws. Purpose-built AI therapist tools that could mislead users about the nature of the relationship are.
How to Use Conversational AI for Mental Health Support Safely
If you’re already using or considering an AI mental health tool, there are practical things that separate helpful use from harmful use.
- Check whether the tool has clinical backing. There’s a difference between Claude and Therabot. One is a general-purpose AI; the other was built on therapy frameworks and tested in an RCT. Know which one you’re using.
- Don’t replace crisis care with AI. If you’re experiencing suicidal thoughts or a mental health emergency, call or text 988 (Suicide and Crisis Lifeline). No conversational AI chatbot is equipped for that.
- Tell your therapist you’re using AI. This is more useful than it sounds. Your AI conversations may surface patterns your therapist hasn’t heard about yet. Bring them in.
- Watch for dependency patterns. If you notice that you’re reaching for the chatbot before calling a friend or family member, that’s worth examining. AI mental health support works best when it supplements, not substitutes, human connection.
- Use it for reflection, not diagnosis. A mental health chatbot is good at helping you organize thoughts, process a difficult day, or prepare for a therapy session. But it’s not equipped to diagnose.
- Read the privacy policy. Especially for apps marketed as AI therapist tools. Data collected during mental health conversations is sensitive. You deserve to know where it goes.
The Future of Conversational AI in Mental Health Care
The long-term trajectory of conversational AI in mental health isn’t really in doubt. The provider shortage isn’t going away. The demand for on-demand emotional support isn’t either. What’s in doubt is whether the tools will get safer and more clinically rigorous fast enough to match the scale of adoption.
NAMI and Beth Israel Deaconess Medical Center are working to benchmark AI chatbot responses systematically, essentially building the evidence infrastructure that the market currently lacks. Therabot’s trial is the first of its kind, but it won’t be the last. If those results replicate, the case for integrating AI mental health tools into first-line care gets significantly stronger.
Generative AI in healthcare is clearly moving toward clinical partnerships. AI that operates under the oversight of human providers – that’s probably the version that survives the regulatory wave. The companion app model, unstructured, unvalidated, and emotionally sticky, is the one states are already moving to restrict.
Final Thoughts
Conversational AI fills a real gap in mental health access. The data on that is clear, and the Therabot trial is harder to dismiss than the skeptics would like. But the gap it fills is specific: between sessions, at 3 a.m., for people who have no other option at that moment.
It is not a replacement for human connection or clinical expertise. Anyone selling it as one is either confused or being dishonest. The most meaningful thing you can do with an AI mental health tool is use it to show up better for the actual humans in your life, including your therapist.
If you’re navigating mental health concerns and want to understand your options, a good place to start is the 988 Suicide and Crisis Lifeline, NAMI’s helpline, or speaking directly with a licensed mental health provider.
FAQs
No. AI mental health chatbots can support and provide between-session help, but they can’t replace licensed clinical care, especially for serious conditions.
It depends on the state. Illinois, Nevada, Maine, and others have banned standalone AI-based therapy. Most laws still allow AI to assist licensed providers administratively.
Clinically validated tools like Therabot (developed at Dartmouth) have the most research backing. General-purpose LLMs like ChatGPT and Claude are not designed for clinical mental health care.
No. If you are in crisis or experiencing suicidal thoughts, contact the 988 Suicide and Crisis Lifeline immediately by calling or texting 988.
Not always. Unlike licensed therapists, general AI platforms often don’t have HIPAA obligations. Read the privacy policy of any AI mental health tool you use before sharing personal information.
