AI Chatbots Don't Reduce Loneliness, UBC Study Finds
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Ruo-Ning Li at the University of British Columbia led a study of roughly 300 students divided into three groups for two weeks: daily chats with a custom chatbot, text exchanges with an assigned stranger, or journaling.
- The experimental chatbot, named Sam, ran on ChatGPT-4o mini and was prompted to embody "the ideal qualities of a perfect roommate — supportive, encouraging, and always available to listen."
- Students who texted real strangers felt significantly less isolated by the trial's end, while those in the chatbot and journaling groups showed no change in loneliness — the chatbot produced no measurable benefit over writing in a journal.
- The researchers concluded: "Alleviating loneliness requires more than the mere simulation of human emotions and care," directly challenging the framing of AI as a viable loneliness intervention.
- Roughly 16% of US adults and 25% of people under 30 have already used AI for companionship, according to surveys cited in the piece, underscoring how many people the study's null result actually applies to.
- The article traces the illusion of chatbot intimacy back to Joseph Weizenbaum's 1960s program ELIZA, whose crude question-rewriting prompts led users — including Weizenbaum's own secretary — to pour out their hearts.
Why it matters: With roughly 16% of US adults and a quarter of under-30s already turning to AI for companionship, the study suggests these interactions deliver an immediate mood lift but do nothing to close the underlying gap of social isolation. Lonely users may feel they have found a cure when the chatbot is functioning more like a temporary distraction.
