AI, Therapy, and What Cannot Be Outsourced
I use AI almost every day. Not for therapy, but for the work around it. It is a remarkable research tool, something like a much faster and more conversational encyclopedia. I use it for editing and enjoy thinking out loud to it regarding theories I am considering. It serves a function similar to a planner and what my notes app used to do except it talks back. When I am stuck on a phrase or trying to clarify a clinical concept for a blog post, I often turn to it. I am not writing as someone unfamiliar with what AI is or skeptical of its usefulness. My husband specializes in AI so I have been around conversations surrounding its use cases and the existential concerns it raises for approximately ten years.
So when I tried a therapy app the other day, I was not encountering AI for the first time. I was specifically curious what it would feel like in a therapeutic context. And in some ways, I found it useful. The psychoeducation was solid. One feature turned learning about your own trauma responses into something interactive, almost game-like, and I could see how that might draw someone in who would otherwise never pick up a workbook.
As I moved through it, something felt off. There was a quality to the experience I can only describe as dissociative. Trauma, by definition, is what happens when an experience overwhelms our capacity to process it. I found myself wondering whether something gets lost when we try to gamify it, or whether the discomfort of sitting with difficult material is not an obstacle to healing but part of it. I do not know if this is the app's particular design or something more inherent to interactive engagement with traumatic material. It is a question I am still sitting with, not a conclusion I have reached.
What I have come to think more clearly is this. There are real things AI can offer in mental health, and they are not trivial. Skills-based modalities like CBT and DBT are, by design, structured and solution-focused. They operate through protocols, worksheets, behavioral experiments, and repetition. An AI can walk you through a thought record. It can teach you distress tolerance skills and remind you to use them. For someone who needs tools and does not have access to a therapist, that is genuinely valuable, and I would not dismiss it. AI can also help people find language for experiences they have not been able to name, and hold a thought in motion long enough for it to clarify. Some of what good therapy does, AI can approximate well.
What it cannot do is be a nervous system.
Traditional talk therapy, particularly the relational and somatic traditions informed by attachment research, operates on a different principle. The treatment is not the information. The treatment is the relationship. A chatbot can generate a sympathetic response. It can reflect your language back to you in ways that feel validating. What it cannot do is meet you in the embodied sense. Whether you call it empathy or mentalization, what is being pointed at is something that requires a body that has known its own suffering and a mind that has been held by other minds. The capacity is not pattern matching. It is lived resonance. The difference is not semantic. It is felt. And it is the felt experience that changes people in this register.
There is a reason we learn certain things better from other human beings. Think about your favorite class in school. For most of us, it was not the subject that made it memorable. It was the teacher. I learned when another person made me feel seen and genuinely excited by the material. I did not learn much from reading my textbook alone at home, though it occasionally helped me memorize a fact, usually only after rereading it seven times. Information can be transmitted through a screen. But something else happens when knowledge passes through a relationship. It lands differently. It stays.
This maps onto therapy more directly than people realize. Psychoeducation delivered by an app might help you understand, intellectually, that your nervous system is stuck in a threat response. That is valuable. But understanding it in the presence of someone who is simultaneously regulating with you, someone whose calm is not performed but genuinely co-created in the room, is a different kind of learning. It is embodied. It changes not just what you know but how your body holds what you know.
So perhaps the question is not whether AI can replace therapy. Perhaps the question is what kind of change you are looking for. If you want to learn about your trauma responses and build a toolkit of coping skills, I think technology can help with that, and there is no shame in starting there. If what you are after is the kind of learning that reorganizes something at the level of your nervous system, the kind that requires being known by another person and finding that the knowing does not destroy the relationship, then you are describing something different. You are describing what happens between two people in a room.
AI will make some forms of mental health support more available to more people, and that matters. But there is a kind of change that requires being met by another person, one who can stay close to difficult material without flinching and without needing it to be different. That meeting is not a feature of therapy. It may be most of what therapy is. No interface can offer it, however sophisticated. If you are looking for it, you are looking for another person willing to sit with you in your hardest material, and to remain steady there.
