Artificial intelligence has arrived in mental health. Millions of people are using chatbots like ChatGPT, Gemini, or specific “AI psychologist” apps to seek emotional support when they feel anxious, sad, or overwhelmed.
The numbers are striking: 5.4 million young people in the United States use AI for psychological advice. 92.7% say they find it useful. And it’s not just teenagers: adults do it too, typing their fears into a text box at three in the morning.
The question we need to ask ourselves isn’t whether technology is advancing. That’s obvious. The real question is: can AI replace a therapist?
And the short answer is no.
But it deserves a longer explanation.
AI can give good advice. And still, it’s not therapy.
Let’s be honest: artificial intelligence has improved a lot. It can generate coherent responses, apparently empathetic, technically correct. It can offer you emotional management strategies, remind you of breathing techniques, validate your experience, even suggest readings or exercises based on cognitive-behavioral therapy.
And that can feel like help. It can be comforting. It can give you the feeling that someone—something—is listening to you.
But there’s something AI cannot do, and it’s precisely what heals: it cannot create a therapeutic bond.
Why the therapeutic relationship is what heals
The therapeutic relationship isn’t a nice extra in the healing process. It’s not an optional add-on or a “plus” that makes the experience more pleasant. It’s the heart of the therapeutic process.
Decades of psychotherapy research confirm it: beyond technique, beyond theoretical approach, what makes the real difference in the change process is the human relationship established between therapist and client.
And that relationship isn’t built with well-chosen words.
It’s built with presence. With gaze. With emotional resonance. With the capacity to be a genuine witness to the other’s suffering. With the trust that emerges when someone accompanies you, without judging you, in your process.
From Gestalt therapy, we speak of contact. Of the real encounter between two people. Of “being there” that cannot be programmed, algorithmized, or simulated. Because the therapeutic bond is not information: it’s relationship. It’s not advice: it’s shared presence.
AI can give you advice. But it cannot be with you. It cannot hold your pain. It cannot look at you and tell you “I see you.” It cannot adjust its response to the subtle nuance of your tone of voice, to the tear that escapes while you speak, to the silence that says more than words.
Healing doesn’t happen in advice. It happens in encounter.
And that, a machine cannot offer.
What are the risks of using AI as a therapist?
Understanding why people turn to AI doesn’t mean we should normalize it without questioning it. The risks are real.
Lack of clinical supervision
There’s no standard that guarantees the advice a person in crisis receives is clinically correct, ethical, or safe. AI has no training. It has no professional ethics. It has no legal responsibility.
And above all: it cannot detect when a situation requires urgent intervention.
What happens if a person with suicidal ideation asks a chatbot for help? What happens if someone with a serious disorder receives inadequate advice that chronifies their distress instead of addressing it?
AI cannot assess risk. It cannot intervene in emergencies. It cannot refer to specialized resources with clinical judgment.
Opacity and biases
We don’t know what information these models were trained with. We don’t know what biases they contain. We don’t know if they perpetuate stereotypes, if they reproduce harmful messages, if they trivialize suffering.
We’re letting millions of vulnerable people receive “advice” from a system whose internal workings are, to a large extent, a black box.
Perceived usefulness versus actual efficacy
That 92.7% of users say it helped them doesn’t mean their mental health has objectively improved.
Subjective perception of help doesn’t equal clinical improvement.
A person may feel temporary relief from receiving a validating response, while simultaneously entering a dynamic that chronifies their distress instead of addressing it at its root.
Illusion of bond
And perhaps the most subtle risk: that we confuse interaction with relationship. That we think because the machine responds “with empathy,” there’s a real encounter happening.
That we normalize relational loneliness by disguising it as digital companionship.
Why do people prefer talking to a machine?
This is the uncomfortable question we need to ask ourselves.
Why does someone prefer asking a chatbot for help rather than a professional or another person?
It’s not just a matter of access. It’s also a matter of trust.
AI offers immediacy. Privacy. Absence of judgment. 24/7 availability. All of these are real advantages.
But they also reveal something uncomfortable about us, the professionals, the institutions, society: we’ve built care systems that aren’t up to the need.
Overburdened systems. Months-long waiting lists. Inaccessible costs. Social stigma. Distrust of professionals. Fear of being judged, medicated, pathologized.
And yes, also a certain structural loneliness: many people don’t have anyone to truly talk to.
AI didn’t appear to solve the problem. It appeared because the problem already existed.
What does this say about us as a society?
That millions of people are asking an algorithm for psychological help isn’t a technological problem.
It’s a symptom of a collapsed care system.
As a therapist, I’m concerned about AI without supervision. But I’m equally—or more—concerned that we’ve reached a point where a person with real suffering finds it easier to access ChatGPT than a consultation with a professional.
We can’t just criticize technology if we’re not willing to look at the hole technology is filling.
We can’t simply say “AI isn’t therapy” if we’re not able to offer real therapy to those who need it.
So what do we do now?
We need urgent research on the real effects of these tools on mental health. Measuring perceptions isn’t enough. We need to know if people improve, worsen, or stagnate.
We need regulation. Not to prohibit, but to establish minimum standards of safety, transparency, and accountability. Generative AI is being used in mental health contexts without any clinical guarantee. That’s not acceptable.
And we need, once and for all, public mental health systems that work. Accessible. Sufficient. Without endless waiting lists. With well-trained and well-paid professionals.
But above all, we need to recover something we’ve been losing: the capacity to be truly present for one another.
To build real bonds. To accompany each other in suffering without rushing, without quick recipes, without seeking immediate solutions to questions that need time.
Because healing is not information. It’s relationship.
AI can inform you. It can give you techniques. It can validate you superficially.
But it cannot create the relational space where real transformation happens.
It cannot offer you genuine presence. It cannot hold the paradox of your suffering. It cannot accompany you in building new meanings from authentic contact.
It cannot look at you. It cannot breathe with you. It cannot be with you in the here and now of your experience.
And that, in the end, is the only thing that heals.
If we don’t want people to turn to machines, we have to offer them something better. And that something better is real people, with real training, with real commitment, with real presence.
Meanwhile, millions of people continue writing their fears into a text box.
And the machine answers.
But it doesn’t look at them. It doesn’t hold them. It doesn’t accompany them.
If you feel you need a real therapeutic space, where there’s presence and genuine bond, we can talk about it. Book an initial consultation call and let’s see if this space can help you.
