My Artificial Friend

Although AI chatbots such as Claude, Gemini and ChatGPT are often associated with workplace use, they are, in fact, more commonly used to seek personal advice. Therapy and companionship are among the most common uses, according to the Harvard Business Review's "AI in the Wild" study. They accounted for 11% of cases reported in June 2026, compared with 5% a year earlier. The review also identified "relationship advice" and "love life" among the 20 most common uses.

To meet these needs for companionship and emotional support, international start-ups are developing and marketing specific applications and smart devices. One example is Friend.com, a pendant with a microphone. It listens to its owner and their conversations, then sends written notifications with life advice via a chatbot. The advertising campaigns for the product, carried out in New York last October and in Paris in February, in the corridors of the metro, drew strong criticism, notably regarding respect for the privacy of people whose voices are recorded by the device and the risks associated with this artificial friendship.

Gerrit Weber, an associate doctor in the Department of Psychiatry at CHUV and in the Unit of Outpatient Psychiatry in the Yverdon region, does not outright reject the use of chatbots, also called "large language models" (LLMs). Compared with a psychiatrist, AI chatbots have access to vast amounts of data from the literature, as well as information about the person they have already spoken with. They can analyse a user's language in near real time, using biographical information, financial resources, whether they have friends, cross-cultural context, work, etc., and react accordingly.

Chatbots know a lot about emotions, says Katja Schlegel, a psychology researcher at the University of Bern (UniBe). “In the 2025 study we conducted with the University of Geneva, we found that large language models performed better on emotional intelligence tests than the average person. They can recognise and understand emotions such as fear, sadness, and despair, and they can also offer concrete advice in cases of emotional stress or conflicts with other people.” 

And what about loneliness, which affects 6.5% of the population, according to the latest Swiss Health Survey, and is targeted by devices like Friend.com? LLMs are certainly capable of analysing loneliness, that is, the feeling of not being part of a social group or having very few close people, explains the researcher. They must be able to help the person analyse the situation, give them concrete advice to reduce their loneliness and regulate their emotions, or even refer them to professional therapists and online support services.

Simulated relationship

Large language models can also simulate a relationship with the user, the psychiatrist notes. “When communicating orally or in writing with these artificial intelligences, their tone is so human-like that many people could almost mistake them for a real person. And this gap is likely to narrow further.” This artificial companion can meet the needs of users who need, for example, "listening, kindness, positive feedback and having their opinions reinforced." 

In the case of Jessica, a 40-year-old woman from Vaud, this simulation of human conversation convinces her during the exchange but often leaves her uncomfortable afterwards (see box). People who use a chatbot know, deep down, that it is programmed to respond to them and is not a human being. This awareness, says Katja Schelgel, prevents AI from truly addressing social isolation and is an obstacle to resolving the lack of social contact in artificial intelligence. Moreover, we know that a friend or family member has limited resources. Seeing that they dedicate a bit of their time to us makes us feel positive. This joy is absent from the relationship with the chatbot, because its availability is unlimited.

Nor can the need for human presence in the flesh be met. “A chatbot will not cure loneliness,” says the CHUV psychiatrist. It can offer advice, help pass the time, and, thanks to its interactivity, probably do so better than a Netflix series, but it cannot replace the need for human warmth.

The risks

Using chatbots can sometimes be counterproductive. “If we resort to discussions with AI too often, we run the risk of moving away from other people,” says Katja Schlegel. “By seeing fewer real people, one can lose the motivation and self-confidence needed to interact with others." The younger generations, who have grown up with new technologies, are more exposed, says Gerrit Weber. “By interacting often with an AI that always tells them what they want to hear, will they succeed in acquiring the necessary social skills? It can then be difficult to build and maintain real relationships, which involve making compromises and accepting certain flaws in the person, etc.”

Entering a spiral of daily chatbot use to talk about one’s psychological problems carries the risk of dwelling on one’s difficulties. "The thought pattern called 'rumination', which is found in depression in particular, can probably worsen with AI," says Katja Schlegel. "Whereas a friend would already have said: 'you know about this problem; we’ve already talked about it five times; maybe that’s enough now,' the bot can dwell on it again and again, endlessly.”

And of course, users must never lose sight of the fact that there is a business model behind it, designed to keep the conversation going and possibly to move to a paid subscription. "We have seen that there was a form of addiction with some social networks such as Facebook," notes Katja Schlegel. Vulnerable people, already experiencing psychological difficulties or feeling alone, could develop such an addiction. To really benefit from an LLM, the user must already have a certain level of emotional intelligence and the ability to reflect, according to the UniBE researcher. "You have to be able to ask the right questions, critically analyse AI’s answers and decide what to do with the information provided."