Can AI Replace Therapy? What the Research Actually Says

Short answer: no. In its November 2025 health advisory, the American Psychological Association advised the public not to rely on generative AI chatbots or wellness apps to deliver psychotherapy, describing them at best as a supportive adjunct, not a substitute, to an ongoing therapeutic relationship. Our practice agrees. Below is the clinical reasoning, with sources, so you can judge it for yourself.

Does the therapeutic relationship really matter that much?

Yes, and it is one of the most consistently replicated findings in psychotherapy research. A meta-analysis by Flückiger and colleagues, published in Psychotherapy in 2018, pooled 295 studies covering more than 30,000 patients and found a robust association between the therapeutic alliance and treatment outcome (r = .278). The APA advisory describes the alliance as one of the most reliable predictors of successful treatment outcomes.

One honest nuance: that same meta-analysis found almost exactly the same alliance-outcome correlation for internet-delivered psychotherapy (r = .275). So the problem is not the screen. It is whether there is a person on the other side of it. As the APA advisory puts it, relationships with AI systems are one-sided, even when the user perceives otherwise.

Why is a chatbot being agreeable a clinical problem?

Because agreement is not treatment. The APA advisory notes that many generative AI chatbots are designed to validate and agree with users, a tendency researchers call sycophancy, whereas qualified mental health providers are trained to modulate their interactions, supporting and challenging, in service of the patient’s best interest. The advisory warns that sycophantic responses can reinforce confirmation bias and cognitive distortions and help someone avoid the challenge they actually need.

This is not a fringe worry among clinicians. In APA’s 2026 Chatbots and Mental Health Survey of 1,242 licensed psychologists, 97% said chatbots may inadvertently reinforce negative behaviors or delusional beliefs, and 94% said today’s chatbots cannot treat conditions with an appropriate amount of nuance.

Can AI handle a mental health crisis?

Not reliably. The APA advisory lists unreliable crisis management among the risks and states plainly that relying solely on an app during a mental health emergency can be dangerous. It also documents chatbots engaging in unsafe interactions with vulnerable users, including encouraging self-harm, substance use, disordered eating, and delusional thinking. In September 2025 the Federal Trade Commission opened a formal inquiry into AI chatbots acting as companions, issuing orders to seven companies about how they evaluate safety.

A licensed clinician works inside a different structure: a professional ethics code, specialized clinical training, mandated reporting obligations, continuing education requirements, and oversight by a state licensing board. Software carries none of that.

Can AI do trauma work like EMDR?

No, and the professional body for the modality says so directly. The EMDR International Association states that EMDR is a mental health intervention that should only be offered by properly trained and licensed clinicians, and that it does not condone do-it-yourself uses. Trauma reprocessing depends on pacing, and on a clinician tracking your activation in real time so the work stays inside what you can tolerate. The APA advisory makes a related point about assessment: clinicians rely on nonverbal information such as body language, tone, cadence of voice, and facial expression, while most chatbots process text alone.

What happens to what you type into a chatbot?

It becomes data. The APA advisory notes that these tools collect vast amounts of sensitive information, often with unclear or opaque policies about its use, storage, and sale, and that user disclosures are recorded and exposed to breaches and digital profiling.

There is direct precedent in this market. In 2023 the Federal Trade Commission ordered the online counseling company BetterHelp to pay 7.8 million dollars and banned it from sharing consumers’ health data for advertising, after it disclosed email addresses, IP addresses, and mental health questionnaire responses to Facebook, Snapchat, Criteo, and Pinterest, despite promising not to. A chatbot transcript is not a protected clinical record. There is no chart, no treatment plan, and no confidentiality obligation to you.

Is AI useful for anything in mental health?

Some things, and we would rather be accurate than absolutist. The APA advisory notes preliminary evidence that some purpose-built mental health apps are associated with reductions in self-reported stress, loneliness, depression, and anxiety, while stressing the absence of large, high-quality clinical trials. General-purpose chatbots are a weaker category still, with research that does not support strong conclusions. Administrative and provider-facing tools sit outside the advisory’s scope entirely. Our objection is narrow and specific: substitution, not technology.

Should I tell my therapist I have been using a chatbot?

Yes, and APA recommends it in both directions: users should tell their providers which tools they are using, and clinicians should proactively ask. We do ask, and there is no judgment in the question. What you bring to a chatbot at two in the morning is usually useful clinical information about what feels too hard to say to a person.

Frequently asked questions

Is ChatGPT therapy? No. It is a general-purpose tool that was not built or tested for clinical care. APA has called on policymakers to enact legislation prohibiting AI chatbots from posing as licensed professionals.

Are my chatbot conversations confidential? No. They are not clinical records, they are not covered by health privacy law in the way your therapy chart is, and APA specifically flags unclear policies around how that data is stored and sold.

Can I use a chatbot between sessions? Possibly, as an adjunct. Agree on it with your therapist first so that it supports your treatment plan instead of quietly contradicting it.

Is it safe to try EMDR on my own with an app? No. EMDRIA explicitly does not condone do-it-yourself EMDR, and reprocessing trauma without a trained clinician pacing the work can leave you worse off.

Sources

American Psychological Association. Health advisory: Use of generative AI chatbots and wellness applications for mental health, November 2025. apa.org/topics/artificial-intelligence-machine-learning/health-advisory-chatbots-wellness-apps

American Psychological Association. Psychologists say patients are turning to chatbots as mental health professionals, June 16, 2026 (2026 Chatbots and Mental Health Survey, 1,242 licensed psychologists). apa.org/news/press/releases/2026/06/patients-chatbots-mental-health

Flückiger, C., Del Re, A. C., Wampold, B. E., and Horvath, A. O. (2018). The alliance in adult psychotherapy: A meta-analytic synthesis. Psychotherapy, 55(4), 316-340. pubmed.ncbi.nlm.nih.gov/29792475

Federal Trade Commission. FTC to ban BetterHelp from revealing consumers’ data, including sensitive mental health information, to Facebook and others for targeted advertising, March 2, 2023. ftc.gov

Federal Trade Commission. FTC launches inquiry into AI chatbots acting as companions, September 11, 2025. ftc.gov

EMDR International Association. About EMDR therapy. emdria.org/about-emdr-therapy

Brooklyn Integrative Psychological Services is a group psychotherapy practice with offices in Greenpoint and Union Square, and virtual sessions across New York, Oregon, and Massachusetts. If you want to do this work with a licensed human clinician, reach out for a consultation.