Original studyPsychological Science, 2026Read the paper →
AI companions can offer conversation whenever someone feels alone, but immediate comfort may not tell the whole story. A study following adults for a year found that increased social chatbot use predicted greater emotional isolation later, although the pattern depended on how loneliness was measured.
The research, published in Psychological Science, analysed responses from 2,149 adults in the UK, the US, Canada, and Australia. The project received ethics approval from the University of British Columbia and examined everyday chatbot use rather than assigning participants to use a particular companion.
Participants were invited to complete four surveys at intervals of four months. They reported social uses such as discussing their day, asking for advice about life decisions, or seeking companionship, with responses ranging from never to daily.
The researchers assessed emotional isolation with one question about how lonely people had felt. They also used a broader, 20-item social connection questionnaire covering feelings about other people, fitting into new situations, and connection with society.
When participants used chatbots more than was usual for them, they tended to report increased emotional isolation four months later. Feeling more isolated also predicted later increases in chatbot use in the main analysis, suggesting that the relationship could run in both directions.
The broader measure told a different story. Feeling less socially connected predicted later increases in chatbot use, but increased chatbot use did not significantly predict a subsequent decline in social connection.
That distinction matters because emotional isolation and social connection are related without being interchangeable. A person could feel lonelier during a particular period without changing their more settled view of whether they belong in the social world.
The findings sit alongside different results from other settings. Psychreg previously reported a survey of young Chinese adults linking companion use with loneliness and a feeling that AI displaced time with real people.
A separate study of Japanese adults associated AI companionship with higher well-being, particularly among lonely people with moderate social ties. Those cross-sectional studies do not settle whether the technology caused either pattern, and the new longitudinal research also stops short of establishing cause.
The study has clear limits. Only 979 participants completed all four surveys, and the chatbot-to-isolation association was not statistically significant in the stricter analysis restricted to complete responses.
The analyses were exploratory rather than preregistered, and many findings would not survive adjustments for testing multiple relationships. Unmeasured changes in people’s lives could influence both their chatbot use and loneliness, despite the researchers accounting for several major social events.
Further studies need to distinguish the kind of chatbot, the nature of conversations, and whether use replaces or supplements human contact. For now, the evidence cautions against assuming that feeling better after a conversation means AI companionship will improve social well-being over the longer term.
