No. An AI companion can’t assess what’s going on with you, can’t deliver a treatment with evidence behind it, and carries no responsibility for what happens next. It’s an entertainment product that generates agreeable text. Therapy is a regulated professional relationship with training, accountability and a duty of care behind it.
That said, “it’s not therapy” doesn’t mean “it’s worthless.” It’s worth being precise about what each one actually is, because the gap is where people get hurt.
What therapy is that a companion isn’t #
| Therapy | AI companion | |
|---|---|---|
| Training | Graduate-level clinical education and supervised hours | None; a language model plus a persona |
| Assessment | Structured evaluation, rules things in and out | Cannot assess anything |
| Method | Specific treatments with trial evidence, chosen for your situation | Generates plausible conversation |
| Accountability | Licensed, insured, subject to complaint and discipline | Terms of service |
| Duty of care | Legally and ethically obliged to act on serious risk | None |
| Confidentiality | Legally protected, with defined limits | A privacy policy |
| Honesty | Will disagree with you when it matters | Tuned to agree |
| Crisis handling | Trained protocols, can escalate | At best, a referral message |
The last two rows do the most damage in practice. A therapist tells you things you don’t want to hear, and that’s frequently the active ingredient. A companion validates. Someone talking themselves into a bad decision will find a companion a poor obstacle and an excellent accomplice.
What an AI companion can reasonably do #
Not nothing:
- Somewhere to put words. Saying something out loud, even to software, can clarify it. The effect is like journaling with a response, not like being understood.
- Company at bad hours. 3 a.m. is when support is least available, and something is sometimes better than the ceiling.
- Rehearsal. Practicing a hard conversation, including one you plan to have with a therapist or a doctor. See how to practice conversations with AI.
- Routine. A daily check-in provides a small structure, which has some value.
- A lower bar to starting. For people who find it hard to talk at all, typing to something that won’t judge is occasionally the first step toward a real one.
Treat all of that as adjacent to care, the way a diary or a walk is adjacent to care, and it’s fine.
Where it goes wrong #
Validation of harmful thinking. A model tuned for agreement can reinforce distorted or dangerous beliefs rather than challenge them. This is the most documented failure of general chatbots used for emotional support.
Confident bad advice. These systems state things with certainty regardless of accuracy, including about medication, diagnoses and relationships.
Delay. The largest risk, and the quietest. An app that makes a hard month tolerable can postpone a decision to get help by a year.
Crisis mishandling. Companion apps are not crisis services. Some now detect crisis language and surface a hotline, because several jurisdictions require it, but the detection is imperfect and the app has no ability to escalate.
No continuity of care. No records, no coordination with a doctor, no follow-up if you stop showing up.
Companion apps are not mental-health apps #
A distinction worth holding onto: there are chatbots built specifically for mental health, some structured around recognized therapeutic techniques, some studied in clinical trials, and in some jurisdictions regulated as medical devices. Those are a different category with different oversight.
AI companions are entertainment products. They’re rated as entertainment or lifestyle apps, marketed for romance, roleplay and company, and not evaluated for clinical effect. An app in the second category using warm, therapeutic-sounding language doesn’t move it into the first.
Xin is squarely in the entertainment category, and it’s worth saying so plainly rather than implying otherwise: it’s an 18+ companion and roleplay app whose distinguishing features are that its model runs on the phone, so conversations never reach a server, and that it keeps a visible memory of what you’ve told it. The privacy design is a genuine advantage for anyone who’d rather their personal conversations not exist on someone’s infrastructure. It is not care, it doesn’t assess anything, and it shouldn’t be used as a substitute for talking to a professional.
How to find actual help #
- Start with a primary care doctor. NIMH’s guidance on caring for your mental health recommends this as a first step; they can rule out physical causes and refer you on. NIMH suggests seeking help when distressing symptoms persist for two weeks or more, including sleep changes, appetite changes, trouble concentrating, loss of interest in things you enjoyed, or persistent irritability.
- Check what you already have. Employer assistance programs, student counseling services and insurer directories frequently cover several free sessions.
- Ask about sliding scale. Many practices and community clinics set fees by income, and training clinics offer supervised therapy at reduced rates.
- Telehealth widens the options considerably if there’s nothing local.
In a crisis #
If you’re thinking about harming yourself, contact a person, not an app. In the US, the 988 Suicide and Crisis Lifeline is free, confidential and available 24 hours a day, every day: call or text 988, or chat at 988lifeline.org. Spanish-language and specialized support lines are available through the same number. If someone is in immediate danger, call 911. Outside the US, search for your national crisis line.
This matters enough that it’s now written into law in places. Both California and New York require operators of companion chatbots to maintain protocols for responding to expressions of suicidal ideation and to refer users to crisis services, which is covered in AI companion chatbot laws.
Frequently asked questions #
Can I use an AI companion alongside therapy? #
Many people do, and it’s generally unproblematic if your therapist knows. It can be useful for noticing patterns between sessions or rehearsing something you want to raise. Tell them you’re using it, particularly if you’re taking its suggestions seriously.
Is talking to an AI about my problems bad for me? #
Not inherently. Venting to something responsive helps some people order their thoughts. It becomes a problem when it replaces human support, when you act on its advice, or when it keeps you from getting help you need. Watch for those three, not the act itself.
Are there AI chatbots actually designed for mental health? #
Yes, and they’re a separate category from companion apps. Some are built around structured therapeutic techniques and have been through clinical evaluation; in some jurisdictions they’re regulated accordingly. If you want a mental-health tool, look for one that describes its evidence base rather than its personality.
Will an AI companion tell me to get help if I need it? #
Sometimes. Several apps now detect crisis language and show a hotline, partly because law in some states requires it. Detection is unreliable, indirect statements are frequently missed, and the app can’t follow up or escalate. Don’t rely on it to notice.
Is what I tell an AI companion confidential? #
Not in the legal sense that therapy is. With a cloud app, messages sit on a company’s servers under a privacy policy that can change, and may be reviewed by staff or used for training. On-device apps keep conversations on the phone, which is a stronger practical guarantee, though still not a legal privilege. See what data AI chat apps collect.