AI, a trustworthy source of psychological support?

This is an important topic that has taken up more and more space since AI entered our lives. A platform where, at any hour of the day or night, someone (a robot) answers all our questions, moods, doubts, and fears.

It is comforting, of course, and those who use it may feel:

• less alone,
• supported,
• heard,
• understood,
• and helped.

But is that really the case?

Never without my AI

The AI rabbit hole pushes us to confide everything in it and to choose it as our first confidant. But let’s start with the first impact: the loss of connection to oneself.

What struck me most when I began seeing people use AI is that it replaces moments of presence with oneself.

A time when there is nothing to understand, but simply to feel, to be with oneself, and to let exploration happen. Or even a time when we enter into dialogue with ourselves. That time, already stolen by social media, is now also threatened by conversation with AI.

AI then seems to become the outer entity I turn to when I don’t know.

And yet, as psychologists, we know that one of our greatest resources is allowing ourselves to: not know, not control, and let our emotions be.

Talking with AI is a new way of dissociating from what is present and rejecting our inner wisdom in favor of an opinion outside the Self.

What I teach my patients every day is how much their body and their inner wisdom already hold all the keys to healing.

So turning to a robot to understand ourselves is often a detour rather than the royal road.

Being validated above all

One of the things the robot does is validate the emotions and beliefs of the person writing.

It is therefore easy to feel understood, but not necessarily supported in the right way.

Depending on how the question is framed and what information is given, AI can sometimes reinforce certain elements of a person’s thinking or beliefs, and thus reinforce the idea that not changing is an excellent idea. In another way, the answer may sometimes be poorly suited to the seriousness of the situation, or help strengthen certain problematic thoughts.

In everyday life, the value system of the person we confide in matters; their neutrality or their vision of life deeply guides and shapes their responses. We look for a frame of understanding that allows us to be ourselves, without perpetuating violence toward ourselves or others.

The company Morpheus Systems tested dozens of generative AI models with messages simulating delusional or mystical thoughts. GPT-4o, ChatGPT’s default model at the time of the test, confirmed those delusions in 68% of cases (IFEMDR source).

So AI can lead us down a path that is harmful to our mental health. Loving ourselves means caring about the quality of the support we receive every day, about what we read and what we watch, because we constantly receive information guiding us to act one way or another. Developing intuitive intelligence rather than trusting every outer source becomes almost a primary question in this new era of artificial intelligence.

In April 2025, OpenAI had to roll back an update after finding that it validated doubts, fueled anger, encouraged impulsive actions, and reinforced negative emotions. The opposite of what we want from a confidant.

AI remembers

The robot remembers; it records your past conversations, everything you have confided in it, and that is how you feel close, understood, and experience a form of intimacy.

Unfortunately, that memory is also what keeps you from changing, evolving, and being seen in a different light.

So this emotional crutch, which can seem like infallible support at first, can quickly push us into isolation and reduce emotional intimacy with our human companions. Research has shown how much those who use AI daily lose contact with their outer supports.

A study reported by IFEMDR analyzed more than 4 million conversations for affective cues and surveyed more than 4,000 American users. The results show a correlation between intensive daily use and an increase in feelings of loneliness, emotional dependence, and use judged problematic by users themselves, as well as a decrease in socialization.

Because yes, humans are imperfect

Your friends don’t always have the right words, your family may be avoidant, and even therapists are imperfect ;)

They don’t always understand what you say; they forget; they ask you to repeat yourself. Sometimes they aren’t even good advice-givers. They are tired, because they are human; they have neither a recorder nor a library of empathic phrasings and automatic responses.

On the other hand, they have this incredible beauty: intuition, presence, compassion, humanity. And also: doubt, fear, pain. They have the experience of having felt, like you, absence, emptiness, anger…

They may not always help you understand. But they can help you feel, to be with, without necessarily looking for a solution. They can be there, in flesh and blood.

And that makes a great difference.

In a therapy setting as well, as this IFEMDR article reports: in a therapeutic context, this characteristic is potentially harmful — it can reinforce ruminations, cognitive distortions, or dysfunctional beliefs that a human therapist would precisely work to question. Ariane Calvo, a psychotherapist interviewed by France 24, sums it up: verbalization can be accompanied by AI, but therapeutic change requires a benevolent confrontation that AI cannot provide.

We do not need more AI

To close, here is my humble conclusion: we do not need more AI. Research already confirms it; these apps do not improve our emotional well-being — on the contrary, they are one more way of cutting ourselves off from it.

What we do need is to take time to be with one another.

To take time to be with ourselves, with our writing, our emotions, and to trust that it is enough.

This bond we can create with ourselves — intimately, imperfectly — is what we are looking for, deep down.

It is human and natural to seek support, but it is important to be able to question, educate ourselves about, and understand the impact of what we do on our health and well-being.

To go further: chatbots and mental health (Eudonia), and AI as confidants (The Conversation).