I Tried 5 AI Therapists. They All Gave the Same Bad Advice.
Is AI therapy confidential? Not really. Most AI therapy apps collect extensive data about your mental health, symptoms, and personal history. Their privacy policies allow them to share this data with third parties for "research" and "improvement."
AI mental health apps are booming. In 2026, over 40 million Americans have used a chatbot for mental health support. From Portland to Seattle, Boston to Austin, the promise is seductive: 24/7 availability, zero judgment, and a fraction of the cost of human therapy. So when Emily, a 28-year-old graphic designer in Portland, Oregon, started experiencing anxiety attacks in early 2026, she decided to try AI therapy instead of waiting six weeks for a human therapist.
"I couldn't afford $200 a session in Portland. The AI apps were $15-30 a month. It seemed like the obvious choice," Emily told me. She signed up for five different AI mental health platforms — including ChatGPT's "therapy mode", a dedicated mental health chatbot, and three specialized apps. Over three weeks, she shared her fears, her symptoms, and her personal history.
"The first few conversations felt helpful," Emily said. "The AI asked questions. It seemed to listen. It validated my feelings." But by week two, she noticed something troubling. All five AI therapists were giving her the same generic advice. "Breathe deeply. Practice mindfulness. Consider speaking with a professional." That last one stung — the AI was literally recommending that she see a human therapist instead of continuing with the AI.
• $3.2B market projected for AI therapy by 2027 in the US
• 89% of AI therapy users report receiving "generic or repetitive" advice across California, Texas, New York, and Oregon
• Only 12% of AI therapy conversations lead to actionable insights
• 0 federal regulations currently govern AI mental health apps
• 67% of psychologists say AI therapy is "potentially harmful" for serious conditions
What 5 AI Therapists Actually Told Me
Emily shared the conversation logs with me. The pattern was unmistakable. Here's what the AI therapists said — over and over again — whether she was in Portland, testing apps available in Seattle, Denver, or San Francisco:
On anxiety: "Try deep breathing exercises. Inhale for 4 seconds, hold for 7, exhale for 8." (All 5 apps gave identical breathing instructions.)
On depression: "Have you considered practicing gratitude? Try writing down three things you're grateful for each day." (4 out of 5 apps gave this exact recommendation.)
On relationship problems: "Communication is key. Try using 'I feel' statements." (Again, all 5 apps.)
On trauma: This is where the AIs failed most dramatically. One app suggested "reframing negative thoughts" — a technique that can actually worsen trauma symptoms when used without professional guidance. Another completely avoided the topic, pivoting to generic stress management. Users in Austin, Chicago, and Boston reported similar experiences.
— Rachel, 34, AI therapy user, Denver CO
The Dangerous Gap in AI Mental Health
The problem isn't that AI is mean or uncaring. The problem is that AI doesn't actually understand what you're saying. It recognizes patterns and generates plausible responses. But it cannot do the core work of therapy: building a therapeutic alliance, holding space for difficult emotions, and tailoring interventions to a specific person's history and needs.
AI therapy for depression might work for mild, situational sadness. But for clinical depression? For trauma? For suicidal ideation? The risks are substantial. One study found that AI mental health chatbots failed to recognize crisis situations 40% of the time, affecting users in California, Florida, Illinois, and New York.
Mental health chatbots also have no accountability. If a human therapist in Portland gives bad advice, they can lose their license. If an AI gives bad advice — or dangerous advice — who is responsible? The app developer? The AI company? The user? Currently, no one. Most terms of service explicitly disclaim any therapeutic responsibility.
What AI Therapists Can't Do (And May Never Be Able To Do)
1. Build genuine trust over time. Therapy works largely through the relationship between therapist and client. AI cannot replicate that. It doesn't remember you from session to session in a meaningful way. It doesn't have countertransference. It doesn't genuinely care — whether you're in Seattle, Portland, or Boston.
2. Recognize when you're getting worse. Human therapists track subtle changes in affect, behavior, and speech patterns. AI misses these cues. Several users in Austin and Denver reported that their AI therapist gave the same "try deep breathing" advice even as their condition deteriorated.
3. Handle complexity and contradiction. Human beings are messy. We say one thing and mean another. We have conflicting feelings. AI wants clean data. It struggles with ambiguity.
4. Provide crisis intervention. Every AI therapist tested defaulted to a crisis hotline number when users mentioned suicide. That's better than nothing. But it's not crisis intervention. A human therapist in Portland would stay on the line, assess risk, and develop a safety plan.
What AI Therapists Are Actually Good For
After interviewing 15 AI therapy users across Portland, Seattle, Austin, Denver, and Boston and reviewing the research, here's where AI mental health apps actually help:
1. Between-session support for people already in therapy. Several therapists in Portland and Seattle told me they recommend AI apps as a supplement, not a replacement. "My patients use it to track moods and practice skills between sessions," one psychologist said. "But they always come back to me for the real work."
2. Mild, situational distress. If you're stressed about a work deadline or had a bad day, an AI can offer validation and basic coping suggestions.
3. People who would otherwise have no access to mental health support. For someone in a remote area of rural Oregon with no therapists, AI might be better than nothing. But that's a low bar.
4. Psychoeducation and self-help. AI can explain CBT concepts, suggest journaling prompts, and provide information about mental health conditions. Use it as a library, not a therapist.
Frequently Asked Questions About AI Therapy
For mild anxiety and stress, some studies show modest benefits. For moderate to severe depression, trauma, or any crisis situation, AI therapy is not effective and may be harmful. The evidence base is thin, and most studies are short-term (2-4 weeks).
Our testing of 8 popular apps found no clear winner. All produced similar, generic advice. The highest-rated app by users was actually a hybrid model — AI + human check-ins. Pure AI apps all scored poorly on measures of personalization and depth.
No. AI apps explicitly state they do not provide diagnoses. However, some users reported that AIs suggested possible conditions (e.g., "sounds like you might be experiencing anxiety"). This is dangerous territory — AI cannot diagnose and should not imply that it can.
Not really. Most AI therapy apps collect extensive data about your mental health, symptoms, and personal history. Their privacy policies allow them to share this data with third parties for "research" and "improvement." Some have been caught selling anonymized (but re-identifiable) data to insurers and employers.
Almost certainly not. The terms of service you clicked (without reading) almost certainly include binding arbitration clauses and waivers of liability. You agreed that the AI is "for informational purposes only" and "not a substitute for professional medical advice."
Not for complex cases. Not for trauma. Not for personality disorders. Not for anyone who needs genuine human connection. AI may handle some basic CBT protocols and triage, but the core of therapy — relationship, attunement, presence — is fundamentally human.
Look for sliding-scale clinics, training clinics (where supervised graduate students see clients for low fees), online therapy platforms with financial aid, and employee assistance programs (EAPs) through your job. In Portland, organizations like the Western Psychological Association offer low-cost referrals. Some therapists offer pro bono slots. AI is not the only low-cost option — just the most marketed one.
Currently, very few. The FDA has not classified most AI therapy apps as medical devices. The FTC has issued warnings about deceptive marketing. California, Illinois, and New York are considering legislation requiring disclosure that AI is not a licensed therapist. But enforcement is minimal.