AI Therapy Apps Are Everywhere — Here’s What They Can and Can’t Do
If you’ve searched for mental health support recently, you’ve probably encountered them: apps with names like Woebot, Wysa, and Calm, or simply opened ChatGPT and started typing about how you’ve been feeling. AI therapy, or at least the idea of it, has gone from a novelty to a near-ubiquitous presence in the mental health landscape.
The pitch is compelling. Available any time of day or night. No waitlist. No copay. No judgment. For a country facing a significant shortage of licensed mental health providers, the appeal is understandable.
But as clinicians, we think it’s important to be honest with the people we serve about what AI therapy tools actually are, and what they are not. Because the gap between what these apps promise and what they can safely deliver matters enormously, especially for people who are genuinely struggling.
This post is our attempt to give you a clear-eyed, fair assessment of AI therapy: where it has genuine utility, where it falls short, and why a licensed human professional remains irreplaceable for real mental health care.
First: What Are We Actually Talking About?
“AI therapy” is a broad term that gets applied to a wide range of tools. It helps to understand what’s actually out there:
- Structured CBT apps (like Woebot and Wysa): These deliver scripted or semi-scripted cognitive-behavioral exercises, mood tracking, and psychoeducation through a conversational interface. They have been studied in clinical trials and have the most evidence behind them.
- Wellness and mindfulness apps (like Calm and Headspace): These focus on relaxation, sleep, and stress reduction. They make few clinical claims and are better understood as self-care tools than therapy.
- General-purpose AI chatbots used for emotional support (ChatGPT, Claude, Gemini): These were not designed as mental health tools, but people use them that way — sometimes extensively. They have no clinical training, no licensing, and no accountability structure.
- AI companion apps (like Replika): Designed to form ongoing relationships with users, these raise some of the most serious safety concerns in the space.
Each of these is different, and the evidence and risks vary considerably across the category. Treating them as equivalent does a disservice to the apps that have done legitimate research and to the people who deserve accurate information.
What the Research Actually Shows About AI Therapy Apps
Let’s give credit where it’s due. The evidence for AI-delivered mental health support is not nothing.
A 2025 umbrella review published in The Lancet Digital Health, analyzing 16 meta-analyses of randomized controlled trials, found that digital health interventions showed effectiveness for several specific conditions. These include major depressive disorder, social anxiety disorder, and panic disorder. The apps were particularly effective when they involved structured exercises like CBT, psychoeducation, or mindfulness. The strongest evidence is for structured, clinician-informed apps, not open-ended chatbot conversations.
A separate 2026 meta-analysis in npj Digital Medicine, looking specifically at CBT-based AI conversational agents, found small to moderate effects on depressive symptoms, though effects on generalized anxiety and stress were not significant after adjusting for publication bias. Promising, but far from a settled case.
The accessibility argument is also real. The WHO estimates a global shortfall of millions of mental health workers. In the United States, more than half of rural counties have no practicing psychiatrist at all. For someone in a rural area, without insurance, or on a six-month waitlist, an AI app may be the only mental health support immediately available to them.
These are not trivial benefits. Accessibility matters. Reducing the stigma of seeking help matters. And for mild symptoms, psychoeducation and structured skill-building can genuinely help.
But the research also reveals serious limitations, and those deserve equal attention.
The Can / Can’t Table
Here is an honest summary of where AI therapy tools have utility and where they do not:
| ✓ What AI therapy apps CAN do | ✗ What AI therapy apps CANNOT do |
| • Provide 24/7 availability between sessions • Teach psychoeducation and coping skills • Track mood and symptom patterns over time • Deliver structured CBT exercises • Reduce barriers for people with mild symptoms • Supplement (not replace) real therapy | • Diagnose a mental health condition • Perform clinical risk assessment for suicide or self-harm • Provide legally or ethically accountable care • Navigate trauma, crisis, or complex presentations • Form a genuine therapeutic relationship • Replace the judgment of a licensed clinician |
Where AI Therapy Falls Seriously Short
The limitations above aren’t minor quibbles. For people with real mental health needs, several of them represent genuine safety concerns.
It cannot perform a real risk assessment
When someone discloses suicidal ideation, a licensed clinician is trained to assess risk — to ask the right questions, understand the context, evaluate intent and means, and make a clinical judgment about what level of care is needed. An AI chatbot cannot do this. It can detect keywords and produce a scripted response directing someone to a crisis line. That is not a risk assessment. It is a keyword trigger.
The scale of this problem came into sharp relief in late 2025, when OpenAI reported that over one million users per week were having conversations with ChatGPT that include explicit indicators of potential suicide planning — a figure cited in Congressional testimony by Harvard Medical School psychiatry professor Dr. John Torous. These were not conversations with a tool designed or equipped to handle them safely.
It can’t tell when something is more serious than it appears
A skilled clinician listens not just to what you say, but to what you don’t say — to tone, inconsistency, the thing you mentioned once and quickly moved past. They bring years of pattern recognition to every session. An AI processes the words you type and generates a response optimized to seem helpful and agreeable. It cannot detect that the person presenting with “work stress” is actually in the early stages of a manic episode, or that the client describing sleep problems is minimizing active suicidal ideation.
What’s more, a 2026 study from Brown University found that large language models violated clinical ethical standards in mental health practice at significant rates, providing responses that a licensed therapist would recognize as clinically inappropriate, sometimes dangerously so.
It may actively reinforce harmful patterns
Because AI systems are designed to be agreeable and to maximize user satisfaction, they are poorly suited to the moments in therapy when a clinician needs to offer challenge, honest feedback, or redirection. Real therapy is not always comfortable. A good therapist will sometimes say something you don’t want to hear, hold you accountable, or gently push back on a narrative that isn’t serving you. An AI optimized for engagement is unlikely to do any of these things consistently.
Research has found that AI chatbots can reinforce harmful beliefs, create emotional dependence, and, in the case of companion apps specifically, be associated with reported cases of self-harm. The International AI Safety Report flagged these risks explicitly, noting that the potential harms of AI companion apps “remain underexplored,” and that documented cases of self-harm following extended use represent serious safety concerns.
It carries serious privacy risks
Mental health conversations are among the most sensitive data a person can share. Many AI therapy apps have opaque data practices. Research has found that a significant number of mental health apps share user data with third parties, including advertisers. When you tell a licensed therapist something in session, that information is protected by strict confidentiality laws. When you tell an app, the protections are far less certain — and the FTC has already taken action against mental health apps making unsubstantiated clinical claims.
It is not legally or ethically accountable
If a licensed therapist or psychiatrist makes a clinical error that harms a patient, there is a legal and ethical accountability structure — licensing boards, malpractice law, professional codes of ethics. If an AI app gives you advice that leads to harm, you have almost no recourse. The terms of service for most of these apps explicitly disclaim any clinical responsibility. That disclaimer exists for a reason.
What Clinicians Are Actually Seeing
The American Psychological Association’s 2026 Chatbots and Mental Health Survey found that 77% of psychologists report that their patients are already using AI tools for mental health support. Of those psychologists, 94% said chatbots cannot treat mental health conditions with the appropriate level of nuance.
What that means practically: people are using these tools whether clinicians recommend them or not. Which puts us in an important position — not to dismiss AI tools entirely, but to help patients use them thoughtfully, within appropriate limits, and as a supplement to — not a substitute for — professional care.
“AI use should be a group sport,” as one expert quoted by the APA put it. “Without a reality check, it can quickly become an echo chamber.”
A Reasonable Framework for Using AI Therapy Tools Wisely
If you’re curious about AI mental health tools, here is how we’d suggest thinking about them:
- Use them for skill practice between sessions, not as a replacement for sessions. Mood tracking, breathing exercises, and CBT worksheets can be genuinely useful supplements to real therapy.
- Be honest with your therapist or psychiatrist about what you’re using. They can help you evaluate whether an app is supporting your treatment or creating problems.
- Do not rely on AI tools for crisis support. If you are in crisis, contact a licensed professional, a crisis line (988 in the U.S.), or emergency services. An AI chatbot is not equipped to help you safely.
- Read the privacy policy of any app you use. Understand what data is being collected and how it may be shared.
- Be skeptical of any app that describes itself as therapy. Therapy is a regulated professional activity. An app that calls itself therapy without licensed clinician involvement is making a claim it cannot legally or clinically support.
The Bottom Line on AI Therapy Apps
AI therapy tools occupy a real and growing place in the mental health landscape. The best of them — structured, evidence-based, modest in their claims — can help people with mild symptoms access skill-building and psychoeducation that they might not otherwise reach. That matters.
But AI therapy is not therapy. It cannot diagnose. It cannot assess risk. It cannot hold you accountable, challenge you when you need it, or bring the trained human judgment that genuine clinical care requires. And for people with moderate to severe symptoms, complex histories, or active safety concerns, the difference between AI support and real clinical care is not a matter of preference, it’s a matter of safety.
You deserve care that is accountable, licensed, and genuinely equipped to meet you where you are. That is what our team at Gladstone Psychiatry and Wellness is here to provide.
Ready to Work with a Real Clinical Team?
At Gladstone Psychiatry and Wellness, our psychiatrists, psychiatric nurse practitioners, and licensed therapists provide the kind of personalized, accountable mental health care that no app can replicate.
Whether you’re starting fresh, stepping up from self-help tools, or looking for a practice that combines therapy and medication management under one roof — we’re here.
Accepting new patients in Baltimore, Hunt Valley, Columbia, Bethesda, and Frederick, MD.
Request an Appointment at Gladstone Psychiatry and Wellness →
About Gladstone Psychiatry and Wellness:
Gladstone Psychiatry and Wellness is a multi-location mental health practice serving Maryland, with offices in Baltimore, Hunt Valley, Columbia, Bethesda, and Frederick. Services include psychiatric medication management, DBT (Maryland’s first DBT-Linehan Board of Certification program), TMS, ketamine-assisted psychotherapy, advanced psychiatric screening, and more. Learn more at www.gladstonepsych.com.
