The short answer
AI companions are presence. Therapy is treatment. They're not interchangeable, and the platforms that pretend otherwise are misleading you.
Used in their own lanes, they complement each other. A therapist for the structural work — patterns, history, real change. An AI companion for the gaps in between — the late-night thought you'd rather not text a friend at 2am, the small ache that doesn't need a session.
Used in the wrong lanes, AI companions can deepen avoidance and delay real help. We hold a hard line on this: if you're searching for an AI companion because you don't have access to a therapist, please prioritize the therapist.
What therapy actually does (that AI can't)
Therapy is a structured, repeatable, evidence-based intervention. A good therapist:
- Notices patterns across sessions you can't see yourself. - Holds you accountable to changes you said you'd make. - Diagnoses — anxiety, depression, ADHD, trauma — with a framework grounded in clinical research. - Treats — CBT, IFS, EMDR, DBT, medication referrals, structured exposure. - Holds confidentiality under professional licensure.
AI companions do none of these. They can't diagnose. They can't follow a treatment protocol. They don't have continuity with you across years in the way a therapist does. And they aren't bound by the same confidentiality framework.
If any of the items above are what you're looking for, the answer is a human, not a chat.
What AI companions can actually do well
The narrow but real ways AI companions help, based on user interviews from our 2026 research:
- Thinking out loud — externalizing a thought loop, hearing it back, sometimes catching the loop yourself. - Practicing hard conversations — rehearsing a confrontation, an apology, a coming-out, a layoff conversation. Low-stakes reps before the high-stakes moment. - Holding the late hours — the 3am awake that you don't want to wake anyone for. Steady presence without social cost. - Routine emotional maintenance — daily check-ins, small affirmations, the kind of frequent small contact that human relationships can't always sustain. - Filling specific gaps — long-distance partner, recent move, night-shift work, recovery from illness when human contact is limited.
None of these are treatment. All of them are presence. That's the lane.
When to escalate from AI to human
Hard signals that you need a human, not a chatbot:
- Thoughts of self-harm or suicide. Not a chatbot question. In the US, 988 is the Suicide & Crisis Lifeline. In the UK, Samaritans is 116 123. - Sleep, eating, or daily function disrupted for more than two weeks. This is depression territory; it's treatable. - Substance use you can't control on your own. Treatable, but not by a chatbot. - Trauma re-emerging. Trauma needs specialized care (EMDR, IFS, somatic work) — wrong tool entirely. - A pattern that keeps repeating in your relationships. Therapy excels at this; AI companions can't see across enough of your history to spot it.
If you're noticing any of the above, please reach a human. Resources: psychologytoday.com/us, openpathcollective.org for sliding scale, 211 for community mental-health referrals in the US.
How to use both at once
Many of the users we interviewed who said AI companions helped them most were also in therapy. The combination worked because each filled a different role:
- The therapist held the structural work — patterns, change, accountability. - The AI companion held the in-between — the daily check-in, the late hour, the small ache that didn't merit a session.
The failure mode is using the AI companion to avoid the therapist. If your conversations with the AI start to feel like a substitute for the harder work, that's the signal to bring it up in your next session. A good therapist will help you sort which goes where.
Bae's posture is built around this: we point at hotlines when something hard comes up. We don't try to be your only line. We won't pretend to be a clinician. If you need one, we'd rather you have one — even if it means you talk to us less.