AI therapy: the problems the risks and what actually works

01Why AI therapy has serious limitations

Psychotherapy is the use of psychological methods, particularly when based on regular personal interaction, to help a person change behavior, increase happiness, and overcome problems. Psychotherapy aims to improve an individual's well-being and mental health, to resolve or mitigate troublesome behaviors, beliefs, compulsions, thoughts, or emotions, and to improve relationships and social skills. Numerous types of psychotherapy have been designed either for individual adults, families, or children and adolescents. Some types of psychotherapy are considered evidence-based for treating diagnosed mental disorders; other types have been criticized as pseudoscience. However, at least to some extent, psychotherapies are supported by neuroscience, tend to have longer-lasting effects than pharmacotherapies, and can improve the effects of pharmacotherapies, including those involving psychedelics. Effective therapy depends on a relationship — the therapeutic alliance between client and therapist that research consistently identifies as one of the strongest predictors of positive outcomes.

AI chatbots cannot form a genuine therapeutic alliance. They can simulate conversation, but they do not understand the person sitting across from them. They cannot read body language, hear the tremor in a voice, or sense what a client is not saying. These nonverbal cues are not peripheral to therapy — they are central to how skilled clinicians assess risk and adjust their approach.

The limitation is not just about empathy, though that matters. It is about clinical reasoning. A therapist integrates context, history, risk factors, and subtle behavioral signals into a dynamic assessment that changes moment to moment. Current AI systems can process text and generate plausible responses, but they lack the contextual understanding that clinical work demands.

02What AI chatbots get wrong about mental health

Mental health is not a text-only domain. Mental health encompasses emotional, psychological, and social well-being, influencing cognition, perception, and behavior. Mental health plays a crucial role in an individual's daily life when managing stress, engaging with others, and contributing to life overall. According to the World Health Organization (WHO), it is a "state of well-being in which the individual realizes their abilities, can cope with the normal stresses of life, can work productively and fruitfully, and can contribute to their community". It likewise determines how an individual handles stress, interpersonal relationships, and decision-making. Mental health includes subjective well-being, perceived self-efficacy, autonomy, competence, intergenerational dependence, and self-actualization of one's intellectual and emotional potential, among others. Mental health also includes emotional, psychological, and social well-being, and it affects how people manage stress and make everyday decisions. Many presentations — psychosis, severe depression, suicidal ideation, trauma responses — involve states that are difficult to assess through text alone.

AI chatbots can give advice that sounds therapeutic but is clinically inappropriate. They may recommend coping strategies that are contraindicated for certain conditions. They may normalize symptoms that warrant professional evaluation. They may fail to recognize when a user is in crisis and needs immediate intervention.

Hallucination — the tendency of large language models to generate confident but false information — is a particular concern. A chatbot might cite a non-existent study, invent a medication interaction, or fabricate a referral resource. In a mental health context, these errors can have severe consequences if a user acts on them.

03The danger of AI replacing human therapists

The most serious risk is not that AI therapy is bad — it is that it becomes a substitute for care that is unavailable or unaffordable. When someone cannot access a human therapist due to cost, waitlists, or geography, an AI chatbot may feel like a reasonable alternative. But it is not equivalent care.

There is also the risk of dependency and delayed treatment. A user who finds comfort in chatting with an AI may postpone seeking professional help, allowing a treatable condition to worsen. The app provides enough relief to delay action, but not enough to actually resolve the underlying problem.

For users with severe mental illness — psychosis, severe depression, active suicidal ideation — AI therapy is not a lower-quality alternative. It is potentially dangerous. These conditions require professional assessment, medication management, and sometimes hospitalization. A chatbot that encourages users to manage these conditions independently could cause direct harm.

04What AI can usefully do in mental health

The risks are real, but so are the opportunities — if the scope is appropriate. Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. It is a field of research in engineering, mathematics, and computer science that develops and studies methods and software that enable machines to perceive their environment and use learning and intelligence to take actions that maximize their chances of achieving defined goals. AI has demonstrated genuine utility in several well-defined roles within mental health care.

AI can provide psychoeducation — explaining conditions, treatments, and coping strategies in accessible language. It can offer guided self-help exercises for mild anxiety or stress, with clear disclaimers that it is not a substitute for therapy. It can serve as a triage tool, helping users determine whether they need professional care and connecting them with resources.

AI-powered monitoring — analyzing smartphone data for signs of mood change, sleep disruption, or social withdrawal — can alert clinicians to early signs of relapse in patients with established diagnoses. These are adjunctive tools that extend the reach of human care, not replacements for it.

05The regulatory and liability questions

The regulatory landscape for AI mental health tools is underdeveloped. Many apps operate in a gray area — they are marketed as wellness tools to avoid medical device regulation, but they are used by consumers as de facto therapy. This regulatory gap means there is no required evidence standard, no mandated safety reporting, and no clear liability when something goes wrong.

If an AI chatbot gives harmful advice, who is responsible? The developer? The platform? The user who chose to follow the advice? The legal framework for AI liability in healthcare is still being built, and mental health presents particularly difficult questions because harm can be hard to attribute and delayed in manifesting.

Regulators are beginning to respond. The FDA has issued guidance on clinical decision support software, and the EU AI Act includes provisions for high-risk AI systems including those used in healthcare. But enforcement lags behind deployment, and many apps operate outside any jurisdiction's active oversight.

06What patients should know about AI therapy

For anyone considering an AI mental health tool, the key question is what the tool is for. Is it a supplement to professional therapy, a stopgap while waiting for care, or a replacement? The answer changes the risk calculus entirely.

Users should read the fine print. Is the app clear about its limitations? Does it provide crisis resources? Does it have a privacy policy that explains what happens to conversation data? Many apps share or sell user data — a particular concern for mental health information, which is among the most sensitive personal data.

Anyone experiencing severe symptoms — persistent depression, anxiety that impairs daily function, thoughts of self-harm, hallucinations — should seek professional care directly. AI tools are not designed for these presentations, and relying on them for severe conditions is not a calculated risk but a dangerous one.

07The future of AI in mental health care

The most likely future is not AI replacing therapists but AI extending their reach. Clinician-supervised AI tools — triage, monitoring, guided exercises, documentation assistance — can help overstretched mental health systems serve more people. The therapist remains the clinical authority; the AI handles tasks that do not require clinical judgment.

Research is also advancing. Large language models trained on clinical data — not just general internet text — may eventually produce tools with genuine clinical reasoning. But these systems will need rigorous evaluation, and the history of AI in medicine suggests that early promise often outpaces proven benefit.

The fundamental tension will remain: mental health care is fundamentally relational, and AI is not a relationship. The tools will improve, the regulation will mature, and the market will grow — but the best use of AI in mental health will always be in service of more human care, not in its place.