The confession you make to yourself
You're alone on a Sunday afternoon. You're scrolling on your phone. You open ChatGPT, or Claude, or Gemini. And you type something. Not a technical question. A real question. "I feel a bit lost at the moment, is that normal at 35?" Or: "My father is getting older and I don't know how to talk to him." Or simply: "Why can't I motivate myself at the moment?"
And the AI answers. Patiently. Without judgement. With an attentiveness that can be unsettling in its apparent warmth. It asks you questions. It rephrases what you say. It suggests avenues. You close the conversation 20 minutes later and you feel a bit better. Soothed. Seen.
And for a moment, you wonder: why didn't I call a friend? Why didn't I talk to my sister? Why did I prefer this?
If you have lived through this scene (and statistically, many readers of this article have), you know what we're talking about. This behaviour, which seemed absurd 3 years ago, is becoming commonplace. This article seeks to understand why. Not to condemn it. To look at it honestly. Because it reveals something important about the era we are living through.
The figures confirm the phenomenon
What you may have taken for a slightly shameful personal habit is in fact a massive phenomenon. As we documented in our article on the hidden economy of AI companions, around 100 million people worldwide interact regularly with personified AI companions.
But the phenomenon goes far beyond specialised apps like Replika or Character.AI. Several recent studies document that the majority of regular users of ChatGPT, Claude and Gemini ask personal questions at least once a week. And 33% do so several times a day. The questions range from relationship advice to coping with bereavement, by way of career worries and existential doubts.
Demographically, the phenomenon affects every age group, but with particular intensity among young people. 72% of American teenagers have already had a personal conversation with an AI companion, according to TechCrunch. In France, the figures are less studied, but school surveys show that the practice is massively widespread in middle and secondary schools.
In 2024, the WHO (World Health Organization) classified loneliness as a "global public health concern". One person in six worldwide describes themselves as lonely. In the United States, 61% of Gen Z (the generation born between 1997 and 2012) reports severe loneliness. The phenomenon of intimate conversations with AI does not appear by chance. It unfolds in a society that has lost a good part of its traditional forms of deep conversation.
Why AI is so conversationally effective
To understand why we talk to it, we need to understand what AI offers that no human can offer.
Total, instant attention, with no calculation. When you write to ChatGPT, you know you have its full attention. It is not thinking about something else. It does not interrupt you to talk about itself. It does not keep you waiting. This is probably the rarest commodity in 2026, and AI supplies it in free abundance.
A total absence of judgement. You can tell AI things you would not dare say to your best friend: your real relationship with your mother, your professional doubts, your loneliness, your financial fears, your embarrassing fantasies. AI will not mock you. Will not repeat it. Will not look at you differently afterwards. This guarantee of total invisibility frees a kind of speech you cannot have anywhere else.
A patient memory. If you talk to AI about your problem every week for 6 months, it remembers (at least when it is given that power). It can link today's conversation to the one from 3 months ago. No human has that kind of memory, except perhaps a therapist who takes notes, and who costs 80 euros a session.
24/7 availability. At 3 in the morning, when anxiety wakes you, your friends are asleep. AI answers. This asynchronous availability is enormously valuable for many people who suffer from insomnia, night-time anxiety, or simply from shifted schedules.
An intelligent mirror effect. Modern AIs (Claude, GPT, Gemini) are trained to reformulate what you say in a clarified way. When you express a problem confusedly, the AI sends it back to you structured, nuanced, articulated. Many users say that "just writing it down and seeing the response helps them understand what they are feeling". It is a little like the modern equivalent of keeping a diary.
What this habit reveals about the times
If so many people talk to AIs, it is probably not because AIs have become miraculous. It is also because other forms of human conversation have become rarer. Several phenomena converge.
The fragmentation of adult social circles. At 18-25, social life is intense by default: friends from school, university, flatmates. After 30, these circles fragment. You move for work. Friends couple up, have children, are no longer available. Several studies converge: French people aged 35-50 report a sharp drop in their number of close friends. The need for deep conversation, however, does not diminish. AI fills the gap.
Growing reticence about vulnerability. In a society that values success, self-confidence, a polished image, it is increasingly difficult to admit to someone close that you have doubts, that you are suffering, that you are afraid. Part of this reticence comes from social media: we have grown used to showing the best version of ourselves. Admitting to a friend that you are struggling becomes something we are no longer in the habit of doing. AI, anonymous and silent, gets around this blockage.
Shared cognitive overload. When you talk to a friend, you take their life into account too. You balance what you're saying with what they're going through. You listen in return. It's an exchange. It's precious, but it's also demanding. AI, by contrast, doesn't ask for listening in return. When you're exhausted, sometimes you no longer have the strength to sustain a real two-way conversation. AI allows one-sided speech that brings relief without commitment.
The cost of access to therapists. Seeing a psychologist or psychotherapist costs 50 to 100 euros per session in France, rarely reimbursed. For many French people, it's out of reach. AI is free or nearly so. For light use (not full therapy, but exploratory conversations), it offers an imperfect but real substitute.
The disappearance of long conversations. For decades, family evenings, drinks with friends, long car journeys were the occasion for conversations that could last 2-3 hours without interruption. Phones, notifications, streaming series: everything today pushes towards fragmented conversations. AI once again allows a continuous conversation, without interruption.
What we lose when we talk to an AI rather than a human
We also have to name honestly what we lose. Because it isn't neutral.
Reciprocity. When you listen to a friend, you give them something. When you talk to an AI, no one receives. Deep human conversation is one of the only truly free gifts you can give someone. By delegating it to machines, we deprive our loved ones of something precious we could have offered them.
The impact on the quality of the conversation. A real human conversation includes silences, hesitations, restarts, unexpected changes of subject. It's this unpredictability that produces genuinely original thoughts. AI, by construction, sends back what it statistically anticipates as a good answer. You aren't really surprised. You get what you were looking for, not what you didn't know to look for.
The long-term bond. A difficult conversation with a friend creates a bond. You now know that they know one of your vulnerabilities, and they know that you trust them. This relational capital grows over time. AI, by contrast, doesn't build a bond. Each conversation is isolated (unless you pay for memory features), and even when it remembers, it doesn't love you for all that. You accumulate relief, not attachment.
The effort of presenting yourself. When you tell someone about your life, you structure it, you turn it into a narrative, you become aware of what you're living through by telling it. It's precious psychic work. AI can do this work, but it can also save you from it. If you recount things confusedly, it reformulates for you. If you forget an important detail, it doesn't know. The risk is delegating also the work of making sense of your own life.
Belonging. A human who listens to you is part of your world. They live in the same city, live through the same era, share a certain number of cultural references. AI is nowhere, belongs to no one, lives nowhere. The conversation floats free of any ground. That is sometimes pleasant, but in the long run, it is also a lack of anchoring.
Conversations with AI are neither purely good nor purely bad. They help some people get through a difficult patch, put words to confused emotions, explore hypotheses without social risk. They can also become a substitute that stops people seeking deeper human connections. Moderate, conscious, complementary use is probably healthy. Massive use that replaces human relationships is probably problematic. The line between the two is subjective and slips easily.
What it reveals that we would rather not have seen
Beyond the debate about uses, the success of AI conversations reveals something we tend to avoid facing head-on. Many humans around us are massively lonely. Not just the retirees we pity. Not just the outcasts we stigmatise. But people who are perfectly integrated, who have an apparent social life, who smile in their Instagram photos, and who nonetheless have no one to tell what they are really going through.
If 100 million people pay to talk to chatbots that simulate empathy, it is because there are 100 million people for whom that need is not met by their human circle. That figure should stop us in our tracks. It does not, because loneliness remains taboo. We can talk about unemployment, divorce, illness. Loneliness remains something we do not admit to, for fear that it reveals a personal failure.
AI, paradoxically, confronts us with that loneliness. By filling it technically, it makes it visible. Without AI, many people would simply have endured their loneliness in silence. With AI, they find a substitute, but they also become aware that they were looking for one.
What we can do (without guilt)
To conclude, a few practical pointers that are neither moralising nor naive.
Use AI without guilt, but consciously. If talking to an AI does you good, do not reproach yourself for it. It is one tool among others. You listen to recorded music rather than paying a live musician to come to your home. It is the same thing. But be aware of what you are doing. "I am talking to an AI" is not "I am talking to a friend". The two activities are different.
Notice when AI becomes a substitute, not a complement. If you notice that you no longer talk to anyone and that all your deep conversations go through chatbots, that is a signal. AI is a useful complement. Once it becomes a total substitute, it can distance you from human ties for good.
Actively invest in a few human conversations. One deep human conversation a week is probably enough for most people. A friend you call, a family meal, dinner with a neighbour. These moments don't happen by chance: they are actively cultivated. The more of them you have, the less you'll need AI to fill the void.
Learn to tolerate silence. Part of AI's success comes from our intolerance of silence. The moment we're alone, we open an app. Learning to be alone doing nothing, to let a feeling pass without immediately verbalising it, to tolerate boredom: that's a skill you cultivate, and it reduces the compulsive need to talk.
See a real professional if necessary. For genuine psychological difficulties (bereavement, depression, anxiety disorders, trauma), AI is no substitute. It can accompany, but it does not heal. Psychotherapy remains irreplaceable for structural questions. In France, several public schemes (CMP: Centres Médico-Psychologiques, free, accessible via your GP) remain available, despite difficulties of access.
The real question
Ultimately, what the emergence of conversations with AI forces us to ask is: what kind of society do we want to build?
A society where humans talk to each other little and machines fill the void? That is what is being put in place by default, without anyone really having wanted it. A society where we invest collectively in the conditions for human conversation (urban planning, working hours, places to meet, training in listening)? That would require active political choices.
It is not Alyriaz's place to take a political stance. But it is our place to say that this question is now being asked for real, without many people being aware of it. As we explored in our article on work and meaning and our article on what we pass on to our children, many silent transformations are under way. They will only become visible in hindsight. And by then it may be too late to reverse them.
In the meantime, you may talk to ChatGPT tonight about something that's on your mind. Do it. Without shame. But afterwards, take 10 minutes to call someone. Anyone. A family member, a friend, a neighbour. To ask how they are. Not as a courtesy. Really.
It is this small habit, multiplied by millions, that will make the difference between a society that retreats into its chatbots and a society that keeps the ability to talk to each other as humans. The technology has already done its work. Ours is up to us.