AI companions, the spread of chatbots with which we create emotional relationships
A study by the European Parliament shows benefits, risks, and challenges of the daily and affective use of GenAI tools
by Camilla Colombo
June 6, 2026
From text messages to traditional chatbots, up to those called in English "AI companions," i.e., virtual avatars powered by artificial intelligence that take on the appearance of a friend. The evolution of the human relationship with technology is reaching a new major turning point that, for many, especially minors and young adults, is already a reality. Based on large language models, these tools are designed and built to engage personal emotional interactions that faithfully mirror human relationships in the offline world, creating a strong attachment. The spread of platforms such as Character.AI and Replika and their use by younger audiences have alarmed the European Parliament, which recently published a study on the topic, highlighting that there are still no specific regulations on this matter. Currently, the regulatory framework is supported by the AI Act, the Digital Services Act (DSA), and the General Data Protection Regulation (GDPR).
The evolution of chatbots
Chatbots certainly did not originate with the AI systems we know today; in fact, they predate even the spread of the Internet: the most famous case is Eliza, developed in the 1960s. The big difference between so-called traditional chatbots and today's AI-based ones can be traced to the evolution of machine and deep learning techniques and large language models, which allowed chatbots to go beyond the instrumental-practical function of providing an answer to a question, mostly in text mode, becoming platforms capable of creating relational and emotional connections thanks to virtual avatars that incorporate vocal and visual features.
Some research conducted in 2025 has highlighted the benefits experienced by those who use AI companions: reduction of the feeling of loneliness; a judgment-free space to express one's feelings; the ability to ask for help and advice for mental health issues; development of relational skills that can also be used in offline life. Added to these are educational and mental development benefits for students taught how to use artificial intelligence, although in this case there are requests from various organizations and institutions to limit access to young people under a certain age (13 or 18 years). No differences emerge in the use of AI chatbots between adults and minors: the purposes are mostly the same—entertainment, curiosity, mental health support, romantic interactions. Precisely on this latter aspect, according to a 2024 European Tech Insights report, 20% of European adults (26% of men, 15% of women) state they have a romantic relationship with their AI chatbot.
The COVID-19 pandemic acted as a trigger for the phenomenon precisely to mitigate the strong sense of loneliness and anxiety experienced while stuck at home. Replika reported a 35% increase in the use of its services, but it was mainly ChatGPT, the generative AI chatbot by OpenAI, released in November 2022, that revolutionized the market: just two months after launch, it already had 100 million active users globally; in September 2025, weekly active users were 700 million; last March, over 159 million people in the European Union reported using it monthly. In comparison, Character.AI, one of the most widespread AI companions, has over 20 million active users each month worldwide. Between 2024 and 2025, there was an exponential growth in the use of AI companions and AI chatbots, with a significant decrease in gender differences, while virtual avatars are most successful among young people between 18 and 35 years old.
Physical and psychological harm
The spread of AI companions, precisely because they are capable of creating emotional bonds with users, also presents risks, as shown by some research on the topic. In particular, the primary risk is the confirmation of one's biases, i.e., the fact that, without a real comparison and dialogue between the user and the machine, the user in relating to the chatbot constantly sees their own ideas and prejudices confirmed. Another critical issue is the development of an emotional dependency on the chatbot one turns to, often to alleviate loneliness, but which, due to the emotional and verbal manipulation techniques with which the AI system is built, can lead to further isolation of the user. Essentially, a self-perpetuating vicious cycle. Other side effects of the careless use of AI chatbots are poor interaction with peers, a decline in language and academic skills, exposure to sexually explicit content unsuitable for minors of a certain age. In extreme cases, it has even led to suicide, as reported by ongoing lawsuits in the US.
The complexity of all these cases lies in proving the responsibility of platforms in causing physical and psychological harm to users, especially the youngest, the most fragile, and the least educated. No less relevant is the issue of the relationship between input and output and between freedom of expression and privacy protection, given that the more one creates an emotional interaction with someone or something, the more one is inclined to share private and very personal information, due to the trust relationship established. But also the spread of dark patterns that push users toward additional subscriptions or memberships and the high environmental cost of using artificial intelligence for these purposes.
The European regulatory framework
The European Union does not yet have specific legislation regulating the use of AI companions, but some useful regulatory tools for consumer protection already exist, such as the GDPR, the AI Act, the DSA, and the introduction of an age verification app for user access to platforms. Restrictions and bans on the use of social networks for minors are spreading in various countries around the world, while in China, a law on AI chatbots will come into effect on July 15. The EU is currently working on the Digital Fairness Act for greater protection of online users.
Reproduction reserved ©
