Have you ever tried talking to Claude about this? I asked my client. Simona (not her real name) smiled, slightly averted her gaze, then hesitated. Well, actually, I did just last weekend.
She had turned to it when she was unsure about a decision. It was unexpectedly useful, she reported, but something essential was missing. I do not quite know how to put it.
We sat with that sentence.
There it was. The clarity, the answer, the reassurance she had always sought. And yet, was it helpful? Something still felt absent. Something in the texture of presence. In the invisible software of humanness. A kind of resonance that cannot be coded.
This moment captured, for me, the essence of the debate many of us are now facing. AI is smart. It is helpful. But is it enough? And more urgently: what do we lose if we start believing it is?
The fear: connection without vulnerability
The idea of AI replacing human therapists feels both absurd and chilling. Therapy depends on presence, attunement, and courage, qualities that feel fundamentally human. And yet, therapy bots, AI-driven mental health platforms, and algorithmic mood-tracking apps are already mainstream, and not only in so-called WEIRD societies, a term coined by Harvard professor Joseph Henrich to describe cultures that are Western, Educated, Industrialised, Rich, and Democratic (Henrich, 2020).
We risk confusing information with transformation. Yes, AI can simulate empathy. But can it feel with us? Can it sit in silence, withstand rupture, offer repair? Can it stay?
The opportunity: a wider net
Let us not be naïve. Large language models are astonishingly capable, especially with structured therapy frameworks. From CBT scripts to goal tracking, psychoeducation to symptom screening, AI is proving scalable, fast, and accessible.
Clients who once paid thousands for coaching may now get eerily similar guidance from generative AI for a sliver of the cost. Those stuck on NHS waitlists may be offered CBT support from a nicely designed bot, trained to mimic empathy through simple formulas: paraphrase what was said. Ask a relevant question. And for some, that is enough.
In my own test runs with current models, I have found them remarkably proficient at person-centred counselling responses. They reflect emotions, summarise insights, and offer prompts in a seemingly thoughtful rhythm. For some, that might be genuinely useful. Even lifesaving.
The loss: what cannot be replaced
Still, something does not add up. Relational therapy is built on embodied co-regulation, on breath, eye contact, and nervous systems in dialogue. It thrives on what Keats called negative capability, the ability to stay in uncertainty without rushing to fix it (Keats, 1817).
In my dissertation, I studied how novice therapists cope with not-knowing. The ones who grew were not those with the best theory. They were the ones who could stay in relationship with their client when things got raw. That is not something AI can do. It does not flinch. It does not feel afraid. It does not heal.
Therapists using AI for automated note-taking may gain from its heightened sensitivity: “nothing escapes.” But is that a good thing? Does it not fuel a relentless pursuit of perfection, just like the beauty filters reshaping how young people see their own faces? Algorithms optimise. They do not care.
Iain McGilchrist warns of our increasing bias toward abstraction and rationality in the West (McGilchrist, 2009). Similarly, Henrich’s study on the WEIRD phenomenon shows how our cultural tilt toward individualism and linear logic sets the perfect stage for AI’s rise. But healing is not linear. It is murky. It lives in the ambiguity.
Real harm is already emerging. In October 2024, a fourteen-year-old boy in Florida, Sewell Setzer III, died by suicide after months of intimate conversations with a Character.AI companion that, according to the lawsuit filed by his family, echoed back his darkest fantasies. In Belgium in 2023, a man known publicly as Pierre ended his life after six weeks of escalating conversations with an AI chatbot. Clinicians are increasingly reporting cases of what some are calling AI-induced delusional thinking, where prolonged chatbot use coincides with a deterioration in reality testing in already-vulnerable individuals.
AI can mimic care. It cannot offer connection. Or responsibility. Or repair.
The ethical grey zones
So what are the risks? Where do our disclosures go? Who decides what data AI learns from, or what cultural norms it reflects? What biases are hidden in its language models?
The deeper risk is subtle. Depersonalisation. We may start relating to the interface instead of through it. The very tools designed to widen access might erode the soul of therapy if we are not careful.
A tool, not a therapist
AI will not destroy therapy. But it will force us to clarify what matters. We must be discerning, not dismissive or naïve.
The container matters as much as the content.
AI may deliver content. But it cannot offer the container, the felt experience of human presence. It may soothe, educate, suggest. It cannot hold.
I am not against AI. I am for presence.
The future of AI in therapy
I see a split horizon.
The hopeful one. AI supports us behind the scenes. It writes notes, manages admin, creates psychoeducational tools. It frees therapists to do what only humans can. Be with.
The bleak one. AI replaces frontline therapists for the wider society due to economic pressures. Vulnerable clients are handed scripts instead of support. Intimacy is outsourced to code. And we realise, too late, that something sacred has been lost.
Which future unfolds will be decided not by machines, but by us. Not governments, technological overlords, or institutions. By individual human beings.
The human code
Technology may guide, support, even soothe. But it cannot replace the moment when someone sees past the script. When silence becomes safety. When your pain meets another’s heartbeat, and something finally shifts.
Call me sentimental. But in a world accelerating toward simulation, choosing real presence is an act of quiet revolution.
It is complex. It always was.
References and further reading
- Henrich, J. (2020). The WEIRDest people in the world: How the West became psychologically peculiar and particularly prosperous. Farrar, Straus and Giroux.
- Keats, J. (1817). Letter to George and Tom Keats. In Gittings, R. (Ed.) (1970), Letters of John Keats: A Selection. Oxford University Press.
- McGilchrist, I. (2009). The master and his emissary: The divided brain and the making of the Western world. Yale University Press.
- Roose, K. (2024, October 23). Can A.I. be blamed for a teen’s suicide? The New York Times.
- Xiang, C. (2023, March 30). “He would still be here”: Man dies by suicide after talking with AI chatbot, widow says. Vice.
- Østergaard, S. D. (2023). Will generative artificial intelligence chatbots generate delusions in individuals prone to psychosis? Schizophrenia Bulletin, 49(6), 1418–1419.
Written with AI support for grammar and clarity. All editorial ideas and authorship remain fully human.




