A Gallup survey of 15,925 people aged 15 to 22 in 16 countries finds one in three has used AI as a companion, yet in-person friends still matter most.

A new survey from Gallup and the Digital Wellness Lab at Boston Children's Hospital suggests artificial intelligence has quietly entered the friendship lives of young people. Published on 17 November, the study draws on 15,925 people aged 15 to 22 in 16 countries, from Japan and Brazil to Kenya and South Africa, surveyed between late May and late August 2026. Across countries, a median of 33% of respondents said they had used AI as a social companion or friend. The share rises to about half among daily AI users and falls to 24% among less frequent users. In the United States the figure was 26%. Respondents who use AI this way had 70% higher adjusted odds of landing in the study's higher-loneliness group.

The researchers present this as a link, not proof that chatbots make anyone lonelier: lonely young people may simply be more likely to turn to them. Human friendship still dominates the picture. A median of 93% have a close friend they mostly see in person, against 37% with a close online-only friend, and 41% of in-person friendships were rated high quality compared with 26% of online-only ones. Among those with online-only friends, 68% said they wanted more time with friends face to face. Scrolling short-form video, which 72% do often or very often, was the only online activity tied to lower wellbeing once other factors were taken into account. Views of AI companions also varied widely, with 45% of users in Australia rating them highly against 12% in Poland.

For families, teachers and young people themselves, the findings point to simple priorities rather than alarm. The factors most strongly associated with how young people rate their lives were offline ones: feeling safe at home, a good relationship with a caregiver and a household's sense of financial comfort. Making room for regular in-person time, through clubs, sport, shared meals or simply meeting up, follows directly from the data, as does keeping an eye on long sessions of short-video scrolling. Talking to a chatbot is not a warning sign on its own, but the results suggest it works best alongside human contact rather than in place of it. The study used online samples in 14 countries, so results there are not fully representative. A second phase with focus groups in four countries is due in early 2027.