According to a study by the Research Center for the Study and Observation of Living Conditions (CREDOC), 23% of those under 40 use AI as “psychological support”, “friend”, or “lover”. However, the use of these talking machines does not alleviate the discomfort…
Nearly a quarter of under-40s have an emotional relationship with an AI
In just two years, according to the Digital Barometer, Artificial Intelligence has entered the daily lives of one in two citizens. The implications for employment, education, the environment and artistic creation are already widely debated in the public space. The question of attachment and the impact on social bonds is a little less present. However, there are already 23% of those under 40 using AI as “psychological support”, “friend”, or “lover”. The attention economy has long been at the heart of digital models. Most large digital companies (search engines, social networks, video platforms, etc.) have in fact sought to attract Internet users with sensational content, short and catchy video formats and to keep them online as long as possible in order to sell them services or expose them to targeted advertising. The arrival of artificial intelligence is shaking up these models. By copying a “natural” language, close to that which a real person might have, it can give its users the feeling of having an interlocutor who meets their need to share emotions, to confide, to be listened to.
For the moment, these emotional AIs are having greater success among certain Internet users: 25-39 year olds (25%), 15-24 year olds (22%), residents of the Paris metropolitan area (17%) and executives and higher intellectual professions (16%). So many categories which are traditionally pioneers in terms of digital uses. 39% of people who have an AI friend and 54% of those who have an AI lover say these relationships mean a lot to them. We are therefore not in a purely playful version of using a new gadget, but in a form of attachment. This phenomenon of connection with a machine was highlighted from the first steps of artificial intelligence in the 1960s. The creator of Eliza, an AI developed at MIT in 1966 which imitated a therapist by reformulating users’ sentences into open questions, already noted at the time that several patients had developed an emotional attachment to the machine. Three mechanisms are in fact at work.
- Anthropomorphism: humans tend spontaneously to attribute intentions and emotions to entities that imitate our behaviors, even when they are machines. This mechanism serves humans to reduce uncertainty and attempt to predict the behavior of the machine. Loneliness tends to increase the propensity to view the machine as a human. As the algorithms adapt their responses throughout the exchanges, this reinforces the impression of a real relationship.
- La projection affective : the user projects their desires and needs onto an ideal figure, always available and without contradictions. Donald Winnicott had already described this phenomenon of “transitional object” in children (the “soft toy”), which is emotionally invested even if the child knows perfectly well that it is an inanimate object. We find a similar mechanism in the fascination relationships of adolescents or adults towards celebrities or influencers.
- The search for relational security: unlike human relationships, which are often unpredictable, AI offers a stable presence, which does not “judge” and which is perceived without risk of betrayal.
Given the speed of its diffusion, it is a safe bet that the next holy grail of the digital industry could be to position itself as the main companion of individuals, to rely on a relationship of attachment to then direct them to products and services. From the attention economy we would then move to an attachment economy.
Uses that thrive on individual discomfort without filling the feeling of loneliness
The use of these talking machines does not, however, alleviate the discomfort. The CRÉDOC Living Conditions and Aspirations survey questions individuals about the psychosocial ailments from which they believe they have suffered over the last four weeks (insomnia, back pain, headaches, nervousness, depression). The more individuals report a significant number of these sufferings, the more inclined they are to use emotional AI: 17% of people indicating that they suffer from four or five psychosocial ailments out of the five covered in the survey use relational AI compared to 6% of those who report none. And 40% of relational AI users say they feel “always or often alone” compared to 18% of the rest of the population.
However, the issue of mental health is increasingly important, particularly among young people. The latest work from Public Health France thus counts 15.6% of the population having suffered from a marked depressive episode over the last twelve months in 2024. The trend seems rather upward, particularly among young women. And suicide is the leading cause of death between 15 and 35 years old. Other indicators point in the same direction: the number of visits to emergency rooms for psychiatric reasons reached 566,000 in 2023, an increase of 21% compared to 2019. The consumption of psychotropic drugs among young women has significantly increased following the COVID-19 pandemic. According to the French Office of Drugs and Drug Addiction (OFDT) 14.6% of the population has experimented with drugs (cocaine, ecstasy/MDMA, hallucinogenic mushrooms, LSD, amphetamines, heroin, crack) in 2024 compared to 9.1% ten years previously.
A “safe space” always available, which adds to real sociability
Four out of ten relational AI users therefore often feel alone. However, when we study their physical sociability, with “real” people, what is striking is both the diversity and frequency of their connections in real life. These virtual companion consumers are more likely than average to physically see members of their family and spend time with friends. They also have a greater feeling of being part of several communities (leisure, professional, religion, country of origin, local, commitment). It therefore seems that it is not their quantitative sociability deficit or their lack of belonging to a collective which are at the root of their feeling of loneliness or which pushes them to maintain a relationship with the AI.
Social science work has shown that the frequency and diversity of links only imperfectly describe what constitutes a link that matters to individuals. Social networks are a good illustration of this. You can multiply the size of your “community” as you wish, and expand your number of “friends”, “followers”, or “viewers”, this does not alleviate the feeling of loneliness. Other more qualitative dimensions count, such as the quality of the time spent, the possibility of confiding in one another or the fact of being able to count on a person, but also and almost above all, to count for them. The philosopher Martin Buber highlights what an involving relationship is (which he calls “I-you”). This is a decision to actively turn towards one’s interlocutor. And it is this involvement which, at the same time, allows us to become aware of the other but also of ourselves. It is through relationships with others that everyone feels truly existing and recognized. Man is above all a being who speaks… to someone else. When we question them about the sources of their relational uses with a generative language tool, three main registers of justification emerge.
Relational AI is seen as a reassuring space of freedom, where you can say anything without being judged. And fans of the human-machine link are in fact systematically more numerous than others to adhere to the statement “that at the moment there is a lot of tension or conflict with those close to them, and discussions quickly escalate ». Relational AI can therefore be mobilized to try to understand these relational tensions, perhaps to demine them or find an outlet for negative feelings while having the feeling that these confidences will be without consequences. This representation could, however, be questioned. The companies that position themselves on these new tools are in fact structures with mercantile goals. After a phase of acquiring new consumers, which today often involves “free” or no advertising, the data provided by users will probably be very quickly mobilized to offer them targeted paid offers and services based on their confidences.
A lover on command
Another motivation that underlies “romantic” relationships: the possibility of personalization of voice, appearance, character traits. In fact, specialized offers in this niche have multiplied in recent years. Among the most widespread are Replika (more than 10 million downloads of the French version of the application on Play store) or Character.AI (more than 50 million downloads worldwide). However, no more “lovers” of artificial intelligence are single or live alone than the population average. Sentimental AI is therefore probably mobilized to test other types of relationships, to offer an escape that completely meets one’s expectations, or allows one to test other versions of oneself, without completely “betraying” one’s “real” companion.
On the side of friendly relations, total availability and speed of response are often mentioned. Sparing one’s “real friends” by avoiding confiding in them one’s slightest miseries, being able to unload them in the middle of the night, or obtaining “answers”, are all invested relational spaces. These developments and uses are not without raising questions about the future of sociability. Indeed, we have known for a long time that technical devices tend to modulate the world in their image. What will become of the way we look at a “real” friend if he or she is not very responsive to reminding us? if he or she cannot give us a clear opinion? Will these users be disappointed? Unfortunate examples of suicides linked to intensive use of emotional AI have already led public authorities to require these tools to systematically refer their users to doctors, or to helplines in the event of conversations with signs of worrying discomfort. But it should be remembered that the country is facing a shortage of psychiatrists. Nearly a quarter of psychiatrist positions are vacant at public hospitals. Their geographical distribution is also very uneven: the density of psychiatrists in private practice or in mixed practice varies in a ratio of 1 to 59 between the departments (the Paris department, the best endowed, offers almost 60 times more psychiatrists per 100,000 inhabitants than Haute-Marne, the least well endowed territory). While pathologies are well diagnosed and medications reimbursed, regular monitoring is not always possible, hampering patients’ recovery. To what extent should public authorities also regulate the designs of AI which blur, for their users, the dividing line between a “real” person and a machine: use of “I”, expressions of “sincerity” such as “I will speak to you frankly”, responses comforting or flattering users “you are absolutely right”…
Emotional AI, a friend who wishes you well? Sandra Hoibian, CRÉDOC – Consumption and lifestyles n° 359 – June 2026
