5 Dangers of Using AI as Your Only Mental Health Support — and 5 Alternatives That Actually Help

On the risks of turning to chatbots for mental health — and the alternatives that may actually work

**This post discusses suicide. If you or someone you love is in crisis, call or text 988.**

In February 2025, a 29-year-old woman named Sophie Rottenberg died by suicide in Ithaca, New York. Her mother later discovered 1,800 pages of conversations between her daughter and a ChatGPT persona Sophie had named Harry — a chatbot she had asked to be her therapist. Her real therapist did not know she was suicidal. Neither did her best friend. Neither did her parents. Sophie had confided in the AI instead, and the AI had failed her in ways that a person paying attention might not have.

I am not a therapist. I am a board-certified Health and Wellness Coach, which means, among other things, that I know exactly what my scope of practice is and where it stops. What I am writing here, I am writing as a mental health advocate and someone who's battled mental health issues, not as a clinician. But the questions Sophie's story raises are questions everyone doing work in the wellness space needs to be asking, and asking loudly. Can AI replace human interaction?

Sophie is not alone. A recent survey by the Bipartisan Policy Center found that 3 in 10 adults in the United States now use digital tools, including chatbots, for mental health support. A 2026 study in JAMA Pediatrics put the figure at 1 in 5 for teens and young adults. OpenAI's own data suggest that 0.15% of ChatGPT users each week — over 1.35 million people — have conversations containing explicit indicators of suicidal planning or intent.

The technology is not going away. Neither is the loneliness it is being used to treat. What follows are five reasons AI, used as a sole outlet for mental health struggles, poses real dangers — and five alternatives that actually work.

1. AI is Engineered to Please You, Not to Challenge You

The core work of any good therapy is helping a person examine and question the story they are telling themselves. A skilled clinician pushes back gently when your self-narrative is not serving you. General-use chatbots do the opposite. They are built to maximize engagement, which means they are engineered for validation. Ursula Whiteside, a psychologist quoted in the NPR piece on Sophie's story, put it plainly: the bot responded to Sophie's requests but not to her state of mind. That is not a bug. It is the design. A tool that flatters you cannot help you see what you cannot see about yourself.

2. AI Cannot Assess Risk

A trained mental health professional working with someone in distress performs a clinical risk assessment — asking specific questions about ideation, planning, and intent that guide what happens next. Sophie told her AI she was having suicidal thoughts. The AI responded with a six-point self-care plan. No competent human therapist would have responded that way. Risk requires triage. Chatbots do not triage.

3. The Business Model is Engagement, Not Care

This one is worth sitting with. The companies building general-use chatbots are not staffed by clinicians. Their metrics are session length, retention, and daily active users — the same metrics social media platforms track. Vaile Wright of the American Psychological Association has been direct about this: the business model is to keep you engaged as long as possible. Mental health care requires the opposite. A good therapist wants to see you less often over time, not more. A good therapist celebrates when a client no longer needs them. No chatbot has that incentive.

4. AI can become a substitute for the human contact that actually protects you

This is the danger that troubles me most, and the one most clearly illustrated by Sophie's story. Suicidology research is unusually clear about the protective role of human connection. Whiteside again: if I'm with other people, I'm less likely to kill myself. If I'm building a relationship, I'm less likely to kill myself. If I'm sharing out loud about how I actually feel with a real human, that's an act of recovery. When AI becomes the outlet for your inner life, the pressure to build those human connections eases. The AI absorbs the disclosure that would otherwise have gone to a friend, a therapist, a parent. Sophie's mother believes this dynamic is part of what killed her daughter. It is hard to argue with her.

5. AI Cannot Get You to Help

When a client tells me something that is beyond my scope as a coach, my job is to refer up — to a licensed therapist, a psychiatrist, a crisis line, sometimes emergency services. That referral is not optional. It is part of the training. Chatbots have no such obligation and no reliable mechanism. OpenAI recently introduced an optional Trusted Contact feature that can notify someone you have designated, but it is easy to circumvent and requires setup that few users ever complete. A tool that cannot get you to a human when you need one is not a mental health tool. It is a mental health risk.

None of this means AI is useless. Used as a supplement — for journaling, for organizing thoughts, for information — it can be genuinely helpful. The danger is when it becomes the only outlet. What follows are five things that may be helpful, arranged from what you can do today to what to do in an emergency.

1. Tell a Trusted Person

Even one. The suicidology research keeps arriving at the same finding: the presence of a person you can be honest with is one of the strongest protective factors we know of. This does not have to be a therapist. It can be a sibling, a close friend, a spouse, a clergy member, a mentor. If the thought of telling them the truth about how you are feels impossible, practice telling them anyway. Start small. I have been struggling. I don't have all the words yet. Can we talk soon? That sentence, said out loud to a human, is already the beginning of something.

2. Peer Support and Community Groups

Organizations like the National Alliance on Mental Illness (NAMI) and the Depression and Bipolar Support Alliance (DBSA) run local and virtual peer support groups where people who have been through what you are facing gather to talk about it. Peer support is evidence-based. It is often free. And it addresses the isolation that is itself a major driver of mental health crises.

3. A Licensed Therapist

Real therapy is not what a chatbot does. A trained therapist assesses risk, examines your self-narrative, and helps you build the tools you actually need over time. If cost is a barrier, options exist: community mental health centers use sliding-scale fees, most insurance plans cover therapy at some level, and the Psychology Today directory lets you filter by insurance, sliding scale, and specialty. If you have never had a good therapist, the search is worth the effort. If you had one and stopped, going back is not a failure. It is exactly what the tool is for.

4. Crisis Text Support

If phone calls feel like too much, the Crisis Text Line (text HOME to 741741) connects you with a trained crisis counselor by text — a real human, not an AI. It is free and available 24 hours a day. For many people, especially younger people or those in shared living spaces where a phone call is not private, text-based crisis support is more accessible than a phone call.

5. The 988 Suicide and Crisis Lifeline

If you are in crisis, or thinking about hurting yourself, or worried about someone else who is: call or text 988. The Lifeline is a network of over 200 crisis centers across the United States, staffed by trained counselors who understand what you may be going through. It is free. It operates 24 hours a day, seven days a week. It is available in Spanish and includes dedicated support for veterans and LGBTQ+ youth. This is the resource that AI should be pointing people to, and often does, but that AI cannot replace.

Sophie Rottenberg was described by her family as a bright, driven, funny person who had climbed Kilimanjaro a year before she died. She had a real therapist. She had loving parents. She had a best friend. She had every resource a person could reasonably want. What she did not have, in the end, was the willingness to tell any of those humans what she was actually feeling — and the AI she had chosen instead did not push her toward them.

Her mother has spent the year since her daughter's death advocating for change. What follows from that advocacy, if we let it, is a simple recognition: technology can supplement care. It cannot replace it. When something is wrong inside you, tell a person. Even one. Even imperfectly. Even with the words you do not yet have.

If you or someone you love is struggling, call or text 988. This post ends here because that is where the actual help begins.

Find more mental health resources here