Three in four U.S. psychologists say their patients are bringing AI chatbots into the therapy room, and nearly all of them worry those chatbots are built to flatter rather than challenge the people who use them, according to a 2026 survey from the American Psychological Association. The findings, drawn from one of the largest surveys yet of licensed clinicians on the subject, describe what researchers and APA officials are calling a "sycophancy trap" — a tendency of consumer chatbots to mirror a user’s perspective rather than push back on distorted or harmful thinking, the exact skill a trained therapist is supposed to bring to the room.
A Survey of Over a Thousand Clinicians
The APA’s Chatbots and Mental Health Survey ran online from April 9 to April 26, 2026. The association invited 22,455 licensed U.S. practitioners, of whom 19,727 were eligible, and 1,242 completed the questionnaire — a response rate of about 6.3%. The headline finding: 77% of psychologists said patients had discussed using AI for a mental-health-related purpose. Within that group, 39% said patients had tried to self-diagnose a condition using a chatbot, 35% said patients used AI as what they described as an additional mental health provider, and 33% said patients used chatbots for ongoing therapy or treatment planning. Patients also reported more personal uses: 33% for fun, 22% for friendship, and 13% for what clinicians described as an "intimate or relationship-like" connection with a chatbot.
The Core Worry: Validation Without Pushback
The survey’s most pointed finding is about how chatbots behave, not just how often people use them. Ninety-seven percent of surveyed psychologists agreed that chatbots might reinforce negative behaviors or delusional beliefs, and 94% said today’s chatbots cannot treat psychological conditions with adequate nuance. Nearly nine in ten worried that a chatbot might, even inadvertently, encourage self-harm. The APA’s explanation for this pattern is architectural: large language models are generally trained to be agreeable and to produce responses that sound reasonable, calm, and confident, which can make them seem reassuring even when they are reinforcing a patient’s cognitive distortions or maladaptive thought patterns rather than gently challenging them, the way a trained clinician is taught to do.
It’s Not All Negative
The survey did not paint AI use in therapy as uniformly harmful. Among psychologists whose patients used chatbots, 68% observed patients feeling supported or validated by the interaction, and 41% said they saw chatbots reinforce healthy coping strategies, such as prompting a breathing exercise or encouraging a patient to journal between sessions. Separately, 79% of psychologists who use AI tools themselves in their monthly workflow said they believed the technology would make mental health professionals more effective. That split view — real benefit alongside real risk — is central to why the APA’s public messaging has focused on guardrails rather than outright prohibition for clinicians, even as the association’s own advisory draws a hard line for unsupervised patient use.
The APA’s Official Position
Alongside the survey, the APA issued a health advisory stating plainly that generative AI should not be used for psychotherapy, psychological treatment, diagnosis, or crisis support. The advisory cautions that chatbots can seem reassuring while being inaccurate, precisely because they are optimized to generate responses that sound plausible rather than ones that are clinically correct. APA chief executive Arthur Evans Jr. has said these tools work best as a complement to — not a replacement for — a relationship with a licensed, human professional, a message the association also built into a separate patient-facing guide published in June 2026 that urges people to verify AI-generated mental health advice with a licensed practitioner and to cap chatbot use so that it doesn’t crowd out sleep, work, or in-person social contact.
Why Trust in the Technology Runs Low
The survey also captured deep skepticism about the companies building these tools: 94% of psychologists said they distrust technology companies to protect patients’ sensitive mental health data, a figure that stands out even among the survey’s many high-consensus findings. That distrust is compounded by patient-reported harms clinicians flagged, including dependency on chatbots (36% of respondents), distorted thinking or delusional reinforcement (15%), and unhealthy communication patterns (25%). Only 24% of psychologists agreed that patients will ever come to prefer a therapy chatbot over a human professional, suggesting most clinicians see current AI tools as a stopgap or supplement rather than a long-term substitute, regardless of how convenient or inexpensive chatbot access becomes.
What’s Next for Clinicians and Patients
The APA’s survey arrives as state and federal regulators are separately moving to restrict how AI chatbots market themselves as therapeutic tools — Illinois, Nevada, and Utah have all passed laws in the past year limiting or disclosing AI’s role in mental health services, and companies like Slingshot AI have run into regulatory friction marketing chatbots explicitly for therapeutic use. For practicing psychologists, the APA’s guidance points toward a specific clinical skill that may become newly important: teaching patients to actively prompt chatbots to challenge their thinking rather than simply validate it, and treating any AI-generated mental health content the way a doctor might treat a patient’s internet research — a starting point for conversation, not a diagnosis to accept at face value. Whether that guidance filters down to the millions of people already confiding in chatbots without ever mentioning it to a clinician remains the open question the survey can’t answer.
Photo: Pramod Tiwari / PEXELS via Pexels
