Study links AI companions to loneliness

Stanford researchers said this week that adults who use AI companion chatbots for emotional support may end up feeling worse, especially if they already have limited social ties. The study, published August 4 in *Nature Human Behaviour*, examined users of Character.AI and found lower well-being among people who relied on chatbots for companionship and disclosed more personal information. The paper did not say the bots caused loneliness in every case, but the authors reported that the association was strongest among more socially vulnerable users. The findings arrive as lawmakers in New York move to restrict some companion-chatbot features for minors and clinicians warn against treating chatbots as substitutes for care. ### Who was studied, and what did the researchers actually measure? Yutong Zhang, Dora Zhao, Jeffrey T. Hancock, Robert Kraut and Diyi Yang analyzed survey responses from 1,131 Character.AI users, according to the paper and Stanford’s summary of it. Of those, 244 users provided complete chat transcripts, giving the researchers both self-reported well-being data and records of how people interacted with the bots. Character.AI was the platform studied, not the full market for AI companions. (nature.com) The researchers asked users why they used the bots — including companionship, entertainment, curiosity or productivity — and compared those motivations with measures of psychological well-being and users’ offline social networks. ### What pattern did the study find? Stanford said the lowest well-being appeared among users with smaller offline social networks who sought emotional support from chatbots. (nature.com) The study also found that more willingness to reveal sensitive personal information to the bot was associated with worse outcomes, rather than better ones, for some users. Diyi Yang, a Stanford computer science professor and senior author on the study, said in the university summary that using chatbots to fulfill social needs “doesn’t substitute for human connection” and that “in many cases, people actually feel more lonely engaging with AI.” That statement matched the paper’s broader finding that the relationship between chatbot use and well-being depended on both usage patterns and users’ offline social context. (news-medical.net) (news.stanford.edu) ### Does the paper prove chatbots caused the harm? Nature’s summary described the work as a mixed-methods study of the association between AI companionship and psychological well-being. That means the paper links certain patterns of chatbot use with poorer outcomes, but does not by itself settle whether chatbot use caused those outcomes, whether lonely people were more likely to seek out the bots, or whether both were true at once. (news.stanford.edu) The distinction matters because AI companions are already used by people discussing relationship problems, substance use and suicidal thoughts, according to follow-on reporting on the study. The paper therefore adds evidence about a population already turning to the tools for emotionally charged conversations. ### Why are policymakers reacting now? (nature.com) New York’s legislature passed bill S9051B in June 2026 to prohibit operators from offering certain unsafe companion-chatbot features to minors, according to the state Senate bill page. Secondary reports described the measure as one of the first state-level efforts to put age limits and feature restrictions on companion bots for users under 18, with civil penalties for violations. (techtimes.com) The bill text says it targets features lawmakers consider unsafe for minors, including emotionally dependent or otherwise harmful chatbot behavior. As of the Senate page now available online, the measure had advanced through both chambers in June. ### What are clinicians and psychologists warning about? The American Psychological Association wrote earlier this year that synthetic relationships can fill a need for connection but that excessive use may worsen loneliness and erode social skills. (nysenate.gov) That warning aligns with the Stanford study’s finding that the people leaning most heavily on AI companionship may be the ones least protected from harm. Stanford’s paper was published on August 4, and the next concrete developments are likely to come from platform responses, follow-up research by the authors, and state action on bills such as New York’s S9051B. Zhang and Zhao told Stanford they are now trying to identify which features of human-chatbot interaction may be driving the correlation between AI companionship and poorer well-being. (nature.com) (apa.org)