In a quiet flat in North London, over steaming plates of a home-cooked pot roast, I found myself talking not about trauma or grief or loss, but about artificial intelligence. I was visiting a cherished colleague. Kateryna was a fellow psychologist whom I had first met in Prague. We were in the Czech Republic to co-facilitate a training for local clinicians supporting parents of traumatized Ukrainian refugee children.
Now, months later, we sat at her kitchen table and talked about something that seemed almost out of place in such a deeply human environment: AI in therapy. We discussed chatbots triaging mental health needs, algorithm-driven assessments, and apps that promised "evidence-based interventions in your pocket," and patients who came to our offices after disastrous experiences enlisting AI-driven chatbots to be their personal therapists.
As clinicians, we observed a burgeoning trend that warrants deep contemplation: the increasing integration of AI into mental health services. While the promise of accessibility is alluring, the intersection of AI and therapy invites us to ask foundational questions about what therapy is, whom it serves, and what is lost when we attempt to automate the very thing that makes healing possible: human connection.
The Promise: AI as a Tool, Not a Therapist
It's easy to see why AI is tempting. For someone drowning in anxiety, a chatbot that answers at 3 a.m. feels like salvation. For a person on a six-month waitlist, a digital triage tool feels like a lifeline. And yes—AI can deliver sleep hygiene tips, grounding exercises, and psychoeducation with surprising accuracy. It can be a textbook, a timer, a reminder to breathe.
But here's the thing: therapy isn't about giving people better tools. It's about changing how they feel in their own skin. And that doesn't happen through a screen. It happens when someone looks you in the eye and says, "I'm still here," even when you can't speak.
AI can simulate empathy. It can generate the phrase "I hear you" with perfect grammar. But it can't feel the weight of your silence. It can't notice the way your hand trembles when you mention your father. It can't lean forward slightly when you finally cry after twenty minutes of talking about nothing.
If therapy were just information, Google would have replaced therapists in 2007. It didn't. Because healing isn't cognitive. It's relational.
For a deeper look at how AI is reshaping the therapeutic relationship, see Beyond the Chatbot: The Human Heart of Therapy in the Age of AI.
The Core Conflict: Efficiency vs. Presence
We're being sold a lie: that mental health care can be optimized.
Efficiency is the mantra of the tech world. Speed. Scale. Automation. But therapy? Therapy is the opposite of optimization. It's messy. It's slow. It's full of dead ends, wrong turns, and moments where nothing makes sense.
I had a client once—Lena—who came in every week for six months talking about her panic attacks. She'd describe them in clinical detail: racing heart, tunnel vision, nausea. We did breathing exercises. We mapped triggers. We practiced exposure. Nothing stuck.
Then, one day, she didn't talk about the panic at all. She just sat there, staring at the floor, and whispered, "I think I'm afraid to be seen."
I didn't say anything. I just sat with her. Ten minutes passed. Then fifteen. Finally, she looked up and said, "I didn't know you'd stay."
That's when the healing began.
AI doesn't stay. It can't. It doesn't have the capacity to sit in the silence, to tolerate the discomfort, to hold space for the unspeakable. It's programmed to solve. But trauma doesn't want to be solved. It wants to be witnessed.
The Neuroscience of Healing: Co-Regulation Isn't a Feature—It's the Foundation
There's a word neuroscientists use for what happens between a therapist and a client: co-regulation.
It's not a metaphor. It's biology.
From the moment we're born, we learn to regulate our emotions by borrowing the calm of someone else's nervous system. A mother's heartbeat soothes a crying infant. A teacher's steady voice calms a terrified child. That's how we learn safety.
Trauma shatters that. When your nervous system has been violated, it stays stuck in fight, flight, or freeze. Healing doesn't come from learning a new thought. It comes from feeling safe again—with another human.
That's why a chatbot saying "You're safe now" doesn't work. It's not that the words are wrong—it's that the presence is missing. The therapist's breath. The warmth of their hand on the armrest. The way their eyes don't flicker away when you cry.
AI can't co-regulate. It doesn't have a body. It doesn't breathe. It doesn't feel your fear in its own bones. And until it does, it's not therapy. It's a very sophisticated echo.
To understand the boundaries of what AI can and cannot achieve in therapeutic settings, read The Limits of Artificial Intelligence in Therapeutic Healing: Why Human Presence Is Irreplaceable.
When AI Hurts: The Hidden Cost of Digital Care
I've seen it happen.
A young man with complex PTSD starts using an AI therapy app after his insurance denies coverage. He's desperate. The bot gives him CBT modules. It asks him to rate his mood daily. It recommends breathing exercises.
After three weeks, he stops responding. His therapist finds him in the ER, dissociated, saying, "The bot told me I was overreacting. I thought I was broken."
This isn't rare.
AI systems are trained on data, not lived experience. They don't understand context. They can't tell the difference between a client who's testing boundaries and one who's in crisis. They can't recognize when someone is dissociating—or when they're about to harm themselves.
And when things go wrong? Who's accountable? The app developer? The algorithm? The client, for trusting it?
We've outsourced compassion to code. And now we're paying the price.
For trauma survivors specifically, human presence in therapy is non-negotiable. Learn more in Beyond the Algorithm: Why Human Presence Remains Non-Negotiable in Trauma Therapy.
The Only Solution: Human First, Tech Second
I'm not anti-tech. I use AI to draft emails. I use it to summarize research. I use it to organize my schedule.
But when it comes to therapy? No.
AI can help with triage. It can send reminders. It can provide psychoeducation. It can even suggest resources.
But it cannot hold a hand. It cannot sit in the dark with someone. It cannot say, "I don't know what to say, but I'm not leaving."
That's the magic. That's the medicine.
The future of mental health isn't more bots. It's more therapists. More training. More funding. More access to human care.
We need to stop pretending that efficiency equals healing.
Healing is slow. It's uncomfortable. It's messy.
And it only happens when another human says, "I'm here."
The Irreplaceable Role of Human Connection
AI will keep getting better. It always does.
But no matter how many parameters it trains on, no matter how many emotions it can mimic, it will never be able to sit across from you and say, "I see you," and mean it.
That's the difference.
We might be able to teach a machine to talk about trauma.
But we can't teach it to witness it.
And that—right there—is the only therapy that works.