The Kitchen Table and the Chatbot
I was visiting a cherished colleague in a quiet flat in North London. We'd met years earlier in Prague, co-facilitating training for clinicians supporting parents of traumatized Ukrainian refugee children. Now, months later, we sat at her kitchen table over steaming plates of a home-cooked pot roast and talked about something that seemed almost out of place in such a deeply human environment: artificial intelligence in therapy.
We discussed chatbots triaging mental health needs. Algorithm-driven assessments. Apps that promised "evidence-based interventions in your pocket." And patients who came to our offices after disastrous experiences enlisting ChatGPT to be their personal therapist.
We kept circling the same worry. Will AI start intruding on the therapy space in ways that weaken the very thing that makes therapy effective? Especially when it comes to trauma, where what is wounded in a relationship can only fully heal in a relationship.
The answer, honestly, is already yes. And the question isn't whether AI has a place in mental health care — it does, in specific and limited ways. The question is whether we're honest about what those limits actually are.
The Scalability Argument
Here's the part where I have to be fair. The global mental health crisis is enormous. The World Health Organization reports that more than a billion people live with a mental health condition worldwide, and treatment gaps remain wide across every income level. Health systems are under-resourced. Waiting lists stretch for months, sometimes years.
In that context, digital self-help tools aren't a luxury. They're a lifeline for people who otherwise get nothing at all.
Take someone struggling with sleep. Ask Gemini how to fix restless nights and you'll get solid, evidence-based advice: consistent bedtime, limit screens before bed, avoid afternoon caffeine. These are the same recommendations I give my own patients. For someone unsure where to begin, that kind of structured guidance is concrete, accessible, and surprisingly helpful. It's easy to implement in exactly the way many of us crave when we're struggling and want a checklist to follow.
The National Institute of Mental Health documents these advantages clearly. Mobile mental health support offers convenience — treatment anytime, anywhere, even at home in the middle of the night or on a bus. It offers anonymity for people who've been reluctant to seek care in person. It lowers cost through free or cheap apps. And it provides 24-hour monitoring or intervention support that no human therapist can match.
The WHO explicitly acknowledges this. They note that the vast care gap for common conditions like depression and anxiety means countries must explore innovative approaches, including digital self-help tools that can be scaled efficiently and affordably.
I'm not dismissing any of this. If an app keeps someone stable until they can get into a therapist's office, that's real value. Don't get me wrong — I'm glad those tools exist.
What the Screen Can't Do
But therapy isn't simply the delivery of information or techniques. If it were, Google and self-help books would have replaced therapists long ago.
Human beings don't change only because they learn new tools. They change because they experience themselves differently in the presence of another person.
This is where things get complicated. The NIMH flags several critical limitations that the scalability argument tends to gloss over: effectiveness, privacy, regulation, and overselling. There is very little industry-wide regulation of mental health apps. Very little information on their actual effectiveness. No consistent standards for evaluating quality. And a real risk that apps promise more than they deliver, turning people away from more effective therapies.
The Psychology Today piece puts it bluntly: AI can approximate empathy in its words, but it cannot simply be with you. It can't sense your subtle facial expressions. Your posture. The tremor in your voice. It can't soften its tone because it notices you withdrawing. It can't be quietly moved by your courage or hold your history in mind from week to week the way a human does, with all the messiness and depth of real memory and care.
It can tell you "you deserve compassion." But it cannot offer you the experience of being met with genuine compassion in real time, from a person whose nervous system is resonating with yours. This limitation is a central concern discussed in The Comfort Trap: Why AI's Fake Empathy Is Dangerous for Your Health, which details how simulated empathy can act as a barrier to authentic healing.
And for trauma work specifically, that lived, bodily experience of safety with another human isn't a nice extra. It's the core of the work.
The Science of Co-Regulation
From infancy onward, we learn to manage distress by borrowing the stability of another person's body, breath, and presence. This isn't poetry. It's neurobiology.
When we've been traumatized — especially in relationships — that regulatory system gets disrupted. Healing often requires practicing co-regulation again: feeling another person stay steady and stay with us while we touch the edges of our pain.
This is why relational trauma must be healed relationally. An algorithm can't detect the micro-shifts in your breathing when you're about to shut down. It can't lean forward slightly because it senses you need encouragement. It can't sit with the silence when words would be premature.
The Psychology Today article makes this distinction crystal clear: AI can deliver structured psychoeducation — sleep hygiene tips, cognitive reframing exercises, grounding techniques. But it cannot mimic the neural safety of real-time co-regulation. Trauma is relational, and it must heal relationally. Linear code simply doesn't have the bandwidth for that.
Think about what actually happens in a trauma session. A patient touches something painful. Their nervous system floods. And across the room, another human stays regulated. Steady. Present. Not fixing anything. Just being there. That presence becomes a template the patient's own nervous system can eventually internalize. Over time, they learn to self-regulate because they've first experienced regulation from someone else.
No chatbot can provide that template. Not because AI is bad. Because it's not alive. In fact, over-relying on chatbots for cognitive and emotional tasks carries risks of its own: The Specter of 'AI Brain': Could Chronic AI Use Cause Computational Brain Injury? outlines how excessive dependency can lead to cognitive offloading and the degradation of active neurological engagement.
Where AI Actually Belongs
So where does that leave us? Am I arguing we should ban mental health apps? Absolutely not.
AI's real place in therapy is as a tool for structured skill practice. Psychoeducation. Sleep hygiene reminders. Mood tracking. Connecting people to peer support or a human provider when the algorithm detects a crisis signal. These are all legitimate, valuable uses.
The NIMH's research into "illness management and supported care" apps is particularly interesting — they're exploring how much human interaction people actually need for app-based treatments to be effective. The answer, so far, seems to be: more than most developers assume.
What I'm arguing for is honesty. Honesty with patients about what a chatbot can and cannot do. Honesty with ourselves as clinicians about not letting convenience replace connection. And honesty with policymakers about the need for real regulation, not just self-certified "evidence-based" labels slapped on an app store listing.
Similarly, the launch of the KAIST Mind Care & Growth Center: A Multidisciplinary Approach to AI-Induced Mental Health Challenges highlights the growing institutional recognition that digital interventions must be rigorously validated through clinical frameworks that prioritize human supervision.
The WHO's Comprehensive Mental Health Action Plan 2013–2030 emphasizes community-based care, cross-sector collaboration, and meaningful investment. Digital tools have a role in that ecosystem — but they're a supplement, not a substitute.
AI will continue to rise. It will shape therapy, healthcare, and daily life in profound ways. But no matter how fast it advances, it cannot replace a steady presence across from you. A regulated nervous system meeting your dysregulated one. The simple, transformative truth of being understood and accompanied by another person.
That's not Luddite nostalgia. That's neuroscience.