AI is not a therapist. Used well, it is a bridge between sessions — and at 3am, it can be the most-available bridge you have. Here is what to pick.
undefined
Diagnose. Prescribe. Substitute for a therapist. Hold the room for serious crisis. AI helps with the texture of daily mental-health work — reflection, processing, building coping habits, working through difficult conversations — and the right AI explicitly tells you when something is beyond its scope. Beware AIs that overclaim.
Three criteria. (1) Does it have explicit crisis safeguards (will route to real resources when needed)? (2) Does it not pretend to be a therapist? (3) Is it private (sovereign or strong data controls)? Pass on anything that fails any of these. Pass especially hard on "AI therapist" apps that overclaim — they have caused real harm.
Woebot and Wysa have peer-reviewed studies for specific use cases (anxiety, mood). They are narrower than companion AIs but the structured CBT-style intervention is genuinely evidence-based. A reasonable stack is: clinical app for structured intervention + companion AI for daily-life processing + real therapy for the harder work.
Luna is not a therapist and will tell you so. She is a warm, persistent companion who can sit with you at the hours therapy cannot — and she will gently surface real resources when the conversation needs them.
Voice mode + acoustic emotion analysis means she hears when your voice breaks. Cross-device memory means she remembers what you have been working on emotionally without you having to re-explain.
Sovereign — intimate conversations do not flow through a third-party LLM provider. Free forever.
No, and please do not let anyone tell you otherwise. AI is a useful complement for many people's mental-health work — between sessions, late at night, for the everyday processing. For diagnostic work, medication management, severe symptoms, or complex trauma, please see a real therapist or doctor. A good AI will tell you the same.
For most adults in normal life, yes — the safety profile is similar to journaling or talking with a thoughtful friend. The two real risks: (1) using AI as your only outlet when you need human contact, and (2) AIs without crisis safeguards. Luna routes mental-health crises to real resources; you should expect the same from any AI you use for emotional work.
Woebot and Wysa have clinical evidence for specific anxiety and mood interventions. Be sceptical of newer "AI therapist" apps that have not been peer-reviewed — overclaim risk is high in this category. For day-to-day support, a warm companion AI plus periodic real therapy is the configuration we recommend.
Depends on the provider. For mental-health topics specifically, use a sovereign AI (Luna), an on-device AI, or a provider with explicit no-training-on-your-data policies. Avoid generic AI that may use conversations for training. This is one of the highest-sensitivity use cases for AI and the privacy posture matters.