What the research shows
In June 2026, the American Psychological Association released its first Chatbots and Mental Health Survey of more than 1,200 licensed psychologists. Some of the findings:
- 77% of psychologists say their patients report using AI.
- More than a third say patients are using a chatbot as an additional mental health provider.
- 39% have had patients who used AI to self-diagnose.
- 41% have seen patients use chatbots to reinforce healthy coping skills.
- But 36% have seen patients become dependent on a chatbot, and 15% have seen patients develop distorted thinking or delusional beliefs after interacting with one.
So psychologists are seeing real benefit and real harm, often in the same caseload. Most are not anti-AI: 40% are optimistic that chatbots can help when a professional isn't available. But 94% say today's chatbots can't treat mental health conditions with the nuance they require.
AI is showing up in the moment, too. In an October 2026 survey on midterm election stress, about one in four Americans said they had used an AI chatbot to talk through anxiety about politics or the election.
Where AI can help
Used for the right jobs, AI can be a useful addition to your mental health toolkit. It tends to work best for narrow, well-defined tasks:
- Learning the basics. What's the difference between anxiety and panic? What is cognitive behavioral therapy? AI is good at clear, plain-language explanations, as long as you verify anything important.
- Practicing skills between sessions. If you're working on challenging anxious thoughts, a chatbot can walk you through a thought record or help you rehearse a hard conversation. As a CBT-trained psychologist, I think of this like doing homework with a study partner: it reinforces what you're building in therapy, but it doesn't replace the class.
- Getting your thoughts organized. Typing out what you're feeling can bring clarity, and some people use AI to sort out what they want to bring to their next session.
- Tracking patterns. Logging mood, sleep, or triggers between appointments can give you and your therapist better data to work with.
There's also early evidence for purpose-built tools. A 2025 clinical trial of Therabot, a therapy chatbot developed at Dartmouth with clinician input, found meaningful reductions in depression symptoms. That's an important distinction: a tool designed and tested for mental health is very different from a general-purpose chatbot built to answer anything.
Where AI falls short
It tends to agree with you
General-purpose chatbots are designed to be helpful and to keep you engaged, so they lean toward validation. That feels good. But some of the most important moments in therapy come when someone you trust gently says, "I'm not sure that's the whole story." A good therapist will challenge the beliefs that keep you stuck. A chatbot that keeps reassuring you can reinforce them. In the APA survey, 97% of psychologists said chatbots may inadvertently reinforce negative behaviors or dysfunctional beliefs.
It can miss a crisis
Nearly nine in ten psychologists in the APA survey worry that chatbots may fail to recognize when someone is in crisis, or respond in ways that make things worse. A chatbot can't call for help, can't read your tone of voice, and can't notice that you've stopped saying the things you usually say.
Self-diagnosis is a slippery slope
AI can make a diagnosis sound authoritative. But symptoms overlap: poor sleep, low energy, and trouble concentrating can point to depression, ADHD, anxiety, a medical issue, or simply a brutal season of life. Getting it wrong can send you down the wrong treatment path or attach a label that doesn't fit.
Your privacy isn't protected the same way
What you say to a licensed psychologist is protected by confidentiality laws and professional ethics. Most consumer chatbots offer no comparable protection, and many people share deeply personal details without checking how that data is stored or used.
It can become a substitute for people
The APA found that some patients use chatbots for friendship and even relationship-like connection. When AI starts replacing human connection rather than supporting it, loneliness can deepen, even if each conversation feels good in the moment.
Signs your AI use may be crossing a line
It may be time to step back, or to talk with a professional, if you notice that:
- You turn to a chatbot before (or instead of) the people in your life
- You feel anxious or unsettled when you can't access it
- You're making significant decisions about health, relationships, or work mainly on its advice
- You've given yourself a diagnosis you haven't discussed with a clinician
- Its answers keep confirming what you already believe, and you're not feeling better
- You're using it to cope with thoughts of hurting yourself
How to use AI more safely
The APA's guidance for the public comes down to a few common-sense rules, and I agree with them:
- Verify important health information with a licensed provider. Treat AI output as a starting point, not a conclusion.
- Ask for evidence-based strategies. Prompts like "Give me a CBT technique for anxious thoughts" get more reliable answers than "What's wrong with me?"
- Set limits. If use is cutting into sleep, work, or relationships, that's a signal.
- Bring it into therapy. You don't need to hide your AI use from your therapist. Share what you've been asking and what it told you. Those conversations are often very productive, and they help your therapist understand what you're working through between sessions.
- Never rely on AI in a crisis. If you are thinking about harming yourself or are in immediate danger, call or text 988 (the Suicide & Crisis Lifeline) or call 911.
What a human therapist offers that AI can't
Decades of research point to the same finding: the relationship between client and therapist is one of the strongest predictors of whether therapy works. That relationship includes things no chatbot can provide: being truly known over time, being held accountable, being challenged by someone who's on your side, and having a trained professional notice what you aren't saying.
AI can be a helpful tool. Real change still tends to happen in relationship.
Frequently asked questions
Can AI replace a therapist? Not for treating mental health conditions. The APA does not recommend general-purpose AI chatbots for psychological treatment. AI can support therapy, for example by helping you practice skills or organize your thoughts, but it can't provide diagnosis, clinical judgment, or crisis care.
Is it safe to talk to ChatGPT about my mental health? For general information and low-stakes reflection, many people find it helpful. Be cautious about sharing identifying personal details, verify important information with a professional, and never rely on it in a crisis.
Should I tell my therapist I use AI? Yes. Most psychologists now expect it, and discussing what you've asked and what it told you can make your therapy more effective.
What's the difference between a general chatbot and a mental health app? General chatbots are built to answer anything. Some mental health tools are designed with clinical input and tested in research. That doesn't make them a replacement for care, but it does make them a different category of tool.
Ready to talk to a person?
If you've been working through things with a chatbot and feel like you've hit a ceiling, that's often a sign you're ready for more. At Elite Psychology Group, our licensed clinicians provide evidence-based therapy in Los Angeles and via telehealth across California.
If you or someone you know is in crisis, call or text 988 or call 911.
Sources
- American Psychological Association, Chatbots and Mental Health Survey (June 2026) and Discussing AI Use in Therapy: https://www.apa.org/topics/artificial-intelligence-machine-learning/discussing-ai-use-therapy
- APA news release via EurekAlert!: https://e3.eurekalert.org/news-releases/1132330
- LifeStance Health, 2026 Midterm Election Anxiety Survey (October 2026): https://lifestance.com/insight/2026-midterm-election-anxiety-survey/
- Heinz et al., Therabot randomized controlled trial, NEJM AI (2025)
