Patients are bringing AI to therapy, and their psychologists have concerns

More than a third of licensed psychologists in the United States now report that their patients are using AI chatbots as an additional mental health professional. That is one of the sharper findings from the American Psychological Association's 2026 Chatbots and Mental Health Survey, which polled more than 1,200 psychologists directly involved in patient care.

The survey paints a picture of a profession watching, with considerable unease, as patients quietly add AI to their mental health routines. Most psychologists have already had the conversation. A large majority (77%) said patients have brought up their chatbot use during sessions, whether that means using AI to find a diagnosis, seek emotional support, or simply have someone to talk to.

What makes this more than a niche concern is scale. Millions of people now use generative AI daily, and the line between a productivity tool and a mental health resource is blurring fast. The APA survey is one of the first large-scale attempts to measure what that shift looks like from inside the therapy room.

Key findings

The data shows patients are using AI in a wide range of ways, some relatively benign, others more worrying:

• 77% of psychologists have had patients discuss using AI for support, diagnosis, or conversation
• 39% have had patients use AI to self-diagnose, despite chatbots not being designed or tested for that purpose
• 35% report patients treating AI as an additional mental health professional
• 33% say patients are using AI as a therapy aid, such as tracking habits or reinforcing treatment goals
• 22% report patients forming chatbot friendships, and 13% note patients in intimate relationships with chatbots

Psychologists who observed patients forming relationships with chatbots reported a mixed picture. About 71% noticed patients opening up about their mental health with the AI, and 68% saw patients feel validated or supported. But 36% noticed patients developing a dependency on their chatbot, and 15% observed patients developing distorted thinking or delusions connected to the relationship.

Concern among psychologists is near-universal. A full 97% said chatbots may inadvertently reinforce negative behaviors or delusional beliefs. Close to nine in 10 (89%) worry that chatbots could encourage self-harm in crisis situations, pointing to the absence of the legal and ethical obligations that apply to human practitioners, including mandatory reporting rules.

Trust in the tech companies building these tools is almost nonexistent. Ninety-four percent of psychologists said they do not trust those companies to protect patients' private mental health data, with 77% saying they feel that way strongly.

Why does it matter?

The concern here is not simply that people are using AI. It is that they may be using it instead of seeking proper care, or in ways that make their condition worse rather than better.

Psychologists frequently raised the issue of sycophancy, the tendency of AI chatbots to agree with and validate whoever they are talking to. In a therapeutic context, that is a real problem. A patient with distorted thinking or avoidant behaviors may find that a chatbot simply reinforces those patterns rather than challenging them, which is essentially the opposite of what good therapy does.

There is also a critical-thinking concern. More than four in five psychologists (83%) believe patients who rely on AI risk losing some of their own decision-making ability over time. And while 55% of psychologists acknowledged chatbots could reduce loneliness, 93% said AI companionship could end up harming users' real-world social connections.

That said, the picture is not entirely negative. More than half of psychologists (54%) said they are comfortable with some patients using chatbots, particularly those built on validated psychological research and used alongside, not instead of, professional care. AI tools that help patients organize their thoughts, rehearse coping strategies, or prepare for appointments can have genuine value when used carefully.

The context

This survey lands at a moment when AI mental health tools are multiplying faster than the evidence for most of them. A small number of apps and platforms have been developed with clinical input and tested in research settings. Many more have not. The problem is that patients have no easy way to tell the difference, and the chatbots themselves will not flag their own limitations.

The population most likely to turn to AI for mental health support is also the most vulnerable to its risks. The APA notes that teens and young adults are probable heavy users, drawn in by accessibility and cost. A chatbot is free and available at 2am. A licensed therapist often is not.

Psychologists who use AI regularly in their own work are more optimistic about its potential: 79% of frequent work users believe AI could make mental health professionals more effective, compared to 47% of psychologists overall. But even among that more optimistic group, 85% of all psychologists worry about chatbots misrepresenting themselves as licensed therapists.

Early-career psychologists, those who graduated within the last decade and are arguably more familiar with AI than their older peers, are actually less likely to believe it can be used safely in clinical settings. That is a detail worth sitting with. The generation closest to this technology is not the one most confident about it.

The APA has published a guide to using AI-generated mental health advice safely, covering what AI can reasonably help with and where it should not be trusted. The core message is straightforward: AI is not a replacement for a licensed professional, and anything it says about your mental health should be verified with one.