Therapists aren’t waiting for research to tell them AI chatbots have entered the therapy room — their patients are telling them directly. A report published by the American Psychological Association in mid-2026, titled “Patients are bringing AI to therapy,” surveyed more than 1,200 licensed U.S. psychologists actively involved in patient care and found the practice has become routine rather than exceptional, even as the same clinicians remain deeply uneasy about it.
The numbers behind a quiet shift in practice
Seventy-seven percent of psychologists surveyed said patients have discussed using AI for mental health support with them. That’s not a passing mention: 39 percent reported patients using AI tools for self-diagnosis, 33 percent said patients use AI to supplement ongoing therapy or treatment, and 35 percent said patients treat a chatbot as something like an additional mental health professional, alongside their actual clinician. The APA has separately noted that roughly 28 percent of adults aged 18 to 29 say they use major consumer chatbots for mental health information or advice — a habit that’s clearly spilling into, and shaping, formal treatment relationships.
What clinicians are actually observing
The picture inside these conversations isn’t uniformly negative. Seventy-one percent of surveyed psychologists said they’d observed patients discussing mental health topics with chatbots in ways that seemed at least somewhat constructive — journaling-style reflection, or working through a thought before bringing it to a session. But 36 percent noticed signs of dependency forming on a chatbot, and 15 percent reported patients showing distorted thinking they linked to chatbot interactions, such as a chatbot reinforcing a belief a human therapist would have gently challenged.
Where psychologist confidence collapses
The report’s starkest numbers are about trust, not usage. Ninety-seven percent of psychologists said they worry chatbots may reinforce negative behaviors or delusions in vulnerable patients — essentially near-universal concern within the profession. Ninety-four percent doubted that chatbots could handle mental health conditions with the nuance a real clinical situation demands, and 89 percent were specifically concerned that a chatbot could inadvertently encourage self-harm, echoing the exact allegation at the center of ongoing lawsuits against major AI chatbot developers. Separately, 94 percent said they distrust tech companies to adequately protect patients’ sensitive mental health data — a concern that predates AI chatbots but that generative tools have sharpened, since a therapy-adjacent conversation with a chatbot may be logged, retained or used to train future models in ways a licensed therapist’s session notes never would be.
Two professions, two very different risk calculations
The gap between what patients are doing and what clinicians trust is itself the story. Patients appear to be treating chatbots as a low-friction, always-available supplement — something to process a feeling with at 2 a.m. between sessions, when a therapist is unavailable and a crisis line feels too extreme a step. Psychologists, by contrast, are trained to spot exactly the failure modes that make them nervous: an AI that validates rather than challenges, that can’t track a patient’s history and countertransference the way a continuing therapeutic relationship does, and that has no license, malpractice liability or clinical supervision behind it if it gets something dangerously wrong. Tanya Carlson, managing director of APA Labs, the association’s technology-focused unit, has argued that setting clear standards matters precisely because failures carry real stakes: “If a product fails, there can be really tragic consequences, so knowing what good looks like is really important.”
Not just an American conversation
The APA’s findings land alongside a fast-moving regulatory environment: five U.S. states restricted AI therapy chatbots in 2026 alone, joining Illinois and Nevada from the prior year, while the EU’s AI Act introduced mandatory disclosure rules in August 2026 requiring conversational AI systems to clearly identify themselves as non-human. None of that regulation, though, directly addresses the scenario the APA’s survey actually documents — a patient using a general-purpose chatbot alongside licensed therapy, a use case that falls outside most of the current state bans, which target chatbots marketed explicitly as therapy or psychotherapy substitutes.
What the profession does with this now
APA Labs says it’s working to translate the survey into practical guidance — for practicing clinicians, on how to ask patients about chatbot use without judgment and integrate it into treatment planning; and for AI developers, on what safety and escalation behavior should look like by default rather than as an afterthought. The open question the report leaves unresolved is enforcement: guidance for clinicians only reaches patients who are already in therapy, while the survey’s own numbers suggest a much larger population is using chatbots for mental health support with no clinician in the loop at all.
Photo: Norman_Gil / PIXABAY via Pixabay
