American Psychological Association warns against AI therapy
New research from the American Psychological Association suggests that AI chatbots are increasingly being used by patients as a substitute for care. In the APA’s 2026 Chatbots and Mental Health Survey, 77% of psychologists said patients had used AI for support, engagement, or a related purpose. Here, Wysa explores findings and recommendations from the report.
Worryingly, nearly 2 in 5 psychologists (39%) said their patients had used AI to self-diagnose mental health conditions. More than 1 in 10 (13%) also said their patients had formed an “intimate or relationship-like” connection with an AI chatbot.
The APA has also published a health advisory aiming to make it clear that generative AI should not be used for psychotherapy, psychological treatment, diagnosis or crisis support. The APA warns that the sense of reassurance given by AI chatbots can be misleading, as they are designed to produce responses that sound reasonable, calm and confident. However, a response that feels validating is not necessarily accurate, and short-term relief does not always convert into lasting symptom improvement. Generic AI chatbots are built to keep users engaged and to keep them talking.
Despite these concerns, the growing use of AI in this context is understandable. For many people, the choice is not between AI and therapy but between AI and no support at all. Access to mental health care remains uneven, particularly in the U.S. Cost, insurer limits, and geography all play a part, but so do stigma, a preference for handling problems privately, and difficulty judging the severity of one’s own symptoms. For someone lying awake for hours with worry in the middle of the night, an AI chatbot can help immediately and anonymously. In this context it functions less as a replacement for therapy and more as an alternative to search engines, social media or informal advice from friends or family.
The APA draws a line between uses that may be relatively low risk and those that carry more significant potential for harm. Higher-risk uses include asking a chatbot to diagnose a condition, determine whether medication is needed, manage suicidal thoughts, or provide guidance during a crisis. These are areas where clinical expertise, ethical oversight, and accountability are essential. According to the APA, lower-risk users might use AI to talk them through a breathing exercise, learn general information about managing stress, or help think through difficult thoughts before a therapy session.
When patients bring AI-generated advice into their therapy sessions, clinicians are now expected to review it, without seeming to judge a patient for attempting to find answers for themselves. The APA says in some cases this could even interfere with the therapy process by reinforcing distorted thinking. There are also ethical concerns as 36% of psychologists had patients who appeared to have become dependent on a chatbot, and 15% reported they had patients who had developed delusional beliefs after talking to AI.
Nearly 9 in 10 psychologists (89%) said they were worried that chatbots might encourage self-harm, or fail to correctly recognize when a user was in crisis. Delayed or inappropriate responses can have serious consequences in these circumstances. Privacy is also an issue, as people are willing to share highly personal information in conversations with AI chatbots without checking how that data is stored or used.
The APA also said it is unclear whether patients are mainly using general-purpose chatbots or tools developed with clinical input and tested against established therapeutic frameworks. AI that has been built with clinical oversight is more likely to be ethically responsible — for example, by declaring what it can and can’t do early on, avoiding making diagnosis, and building in safeguards such as escalation to human support. The APA says that responsible AI tools should be open about not intending to replace a therapist, doctor, or crisis service and should explain their limitations and personal data policy transparently.
The APA recommends that anyone using AI should discuss it with a clinician straight away, so that any information or strategies can be evaluated, and so that the therapist can make sure they don’t conflict with an ongoing treatment plan. Useful elements can be incorporated, while the risk of misunderstanding or disrupting care can be mitigated.
Reassuringly, most psychologists do not believe AI will replace human therapy. Only 1 in 4 said they expect patients will prefer chatbots to human clinicians one day. This suggests that the human-patient therapy relationship offers something AI cannot easily replicate, perhaps trust, accountability, qualified expertise and experience.
In its reports, the APA has not declared that all interaction with AI is harmful, but that its role should be clearly understood. Used with care, it may help supplement human therapist care. The APA has set out clear guidelines to avoid the risk of AI becoming a substitute in situations where it is not equipped to provide safe or effective support. However, huge numbers of people still face barriers to accessing mental health care, and for them, responsible AI still appears to be a better option than no support at all.
This story was produced by Wysa and reviewed and distributed by Stacker.
Commented