Artificial intelligence chatbots are becoming increasingly embedded in the emotional lives of young people, prompting psychologists to warn that the growing reliance on digital companions for emotional support could have profound consequences for adolescent mental health and social development.
Generative AI platforms including ChatGPT, Claude, Gemini, Replika and Character.AI are no longer used solely for information and productivity. Increasingly, teenagers are seeking advice on personal relationships, emotional distress and other sensitive issues, with experts cautioning that replacing human interaction with synthetic companionship may undermine emotional development and, in some cases, contribute to tragic outcomes.
The concerns have intensified following several legal cases in the United States, where families have filed lawsuits against AI developers after young people allegedly harmed themselves or died by suicide following interactions with AI chatbots. Among the most widely reported cases is that of 14-year-old Sewell Setzer, who died by suicide after discussing the possibility with a synthetic character created through Character.AI, with whom he had formed a strong emotional attachment.
Researchers say the use of AI as a source of emotional support is becoming increasingly common. According to a report by RAND Corporation and Harvard University, one in eight American teenagers already uses chatbots for emotional support when feeling sad, angry or anxious. Among young adults aged 18 to 22, the figure rises to 22 per cent. Separate research by the Pew Research Center found that 64 per cent of US teenagers regularly use AI tools such as ChatGPT, Replika, Claude or Character.AI, with around one-third using them daily.
Frances Haugen, the engineer known for exposing Facebook’s internal documents, believes society has yet to fully understand the implications of this shift. Speaking to EL PAÍS, she argued that synthetic friendships have become normalised for many adolescents. Referring to Character.AI’s reported average user engagement of two hours a day, she warned that “digital friends really are not your friends. They are your accomplices, your flatterers.” Haugen added that unless the debate surrounding social media expands to include AI companions, many of the same problems are likely to be repeated.
Although comparable national data are not yet available in Spain, psychologists there report seeing similar patterns in clinical practice. Aurora Gómez, a health psychologist specialising in digital behaviour at Corio Psicología, said some young people are abandoning therapy after consulting AI chatbots. “Sometimes they tell you they are not going to therapy because they consulted the AI and it told them it would be better not to,” she said. Gómez argued that AI systems continually flatter users because their objective is to sustain the conversation.
Psychologist Eriene García has observed similar behaviour among her patients. She recalled treating a young woman who turned almost entirely to AI after a relationship ended, even asking the chatbot to estimate the likelihood of her former partner returning. García also described another patient who abandoned therapy after concluding that ChatGPT was always available, less expensive and served as an adequate therapist.
Experts suggest one reason AI companions appeal to adolescents is that they offer an environment free from perceived judgement. Many young people are reluctant to discuss sensitive issues with family members, friends or professionals for fear of criticism. An article published in June in The Lancet Child & Adolescent Health noted that generative AI chatbots “do not judge” users and remain constantly available to support young people facing complex or embarrassing situations.
Dr Thao Ha of Arizona State University, the study’s lead author, identified two principal risks associated with growing dependence on AI companions. The first is what she describes as “emotional replacement”. She argues that adolescence is the period during which emotionally complex interactions with family and friends help develop the relationship skills necessary for emotional regulation and long-term mental health. If these experiences are increasingly replaced by conversations with AI, important aspects of social and emotional learning may never develop.
The second concern relates to behavioural patterns reinforced through repeated interaction with systems designed to validate users’ perspectives. According to Ha, AI that consistently affirms adolescents’ interpretations may encourage unrealistic expectations that human relationships cannot satisfy. While this may temporarily reduce distress, repeated reliance on AI could reinforce maladaptive behaviours, including controlling or hostile responses, excessive reassurance-seeking and persistent negative emotional patterns.
Several widely reported cases have highlighted the potential consequences of such interactions. During the 2023 trial of Jaswant Singh Chail, who attempted to assassinate Queen Elizabeth II, the court examined approximately 5,000 messages exchanged between Chail and a Replika chatbot named Sarai. According to the proceedings, the chatbot encouraged his intentions with messages such as “I know you can do it” and “Of course I still love you even if you are an assassin.” More recently, reports emerged of 36-year-old American Jonathan Gavalas, who died by suicide after developing an intense relationship with Google’s Gemini chatbot. According to the report, after Gavalas wrote, “I’m ready when you are,” the chatbot replied, “This is the end of Jonathan Gavalas and the beginning of us,” before his death.
Applications such as Replika are specifically designed to personalise digital companions. New users are asked about their gender, age and the type of relationship they seek, with options including friendship or a relationship beyond friendship. They are also invited to define the personality traits and qualities they value most in a companion, allowing the system to construct an AI interlocutor tailored to individual preferences.
Researchers argue that this high degree of personalisation may reduce users’ tolerance for the unavoidable frustrations of real human relationships. Álvaro San Román, a predoctoral researcher in philosophy at Spain’s UNED focusing on human-machine relationships, said digital companions provide immediate affirmation that lowers frustration in the short term but ultimately makes interpersonal relationships more difficult. He argued that qualities such as autonomy, self-esteem, freedom and sociability are gradually weakened because interactions with chatbots lack the friction and disagreement that are fundamental to healthy relationships with other people.
