A chatbot can always be available, patient and ready to reply. Those qualities help explain the appeal of AI companionship—and make the question of what it does to human relationships increasingly urgent.
In a 2 October 2026 interview with Vox, University of North Carolina researcher Anne Maheux discussed a longitudinal study of about 2,300 middle-school students. The report describes an association between greater use of AI for emotional companionship and higher loneliness. The paper was still under review. That finding comes here through the researcher’s interview, rather than our independent examination of the study’s full dataset.
A warning signal needs careful interpretation
“Longitudinal” means following people over time. That can reveal how patterns change and help investigate which changes occur first. It does not automatically remove every alternative explanation.
Loneliness, existing friendship difficulties and the reasons someone turns to a chatbot may influence one another. An observational association does not establish that an AI companion caused loneliness, or show that every young user has the same experience. Those limitations are central to interpreting this early report.
An AI companion is a chatbot used as a social or emotional relationship partner. That use deserves separate attention from asking a system to explain a maths problem. Treating every activity as a single quantity of “AI use” can hide what people are actually doing.
Published research points to a wider evidence gap
A separate peer-reviewed study published on 25 September in AI & SOCIETY gathered perspectives from 141 researchers and adjacent professionals, including 15 interviews. Participants identified relational uses, overreliance and misinformation among research priorities. They also described limited industry transparency and the pace of product change as barriers.
This was a study of expert perspectives, not a trial measuring harm in teenage chatbot users. Its value is in identifying what should be investigated and the practical obstacles to doing that well.
In a 15 September essay for the Association for Psychological Science, Maheux argues that adolescent development involves learning through effort, relationships and self-reflection. She and Chelly Maes examine how generative AI could alter those experiences. This is a developmental argument and research agenda, not proof that every use obstructs growing up.
The design debate cannot wait for perfect certainty
The strongest case for companions is that an accessible conversation may offer comfort when a person feels isolated. The strongest concern is that comfort could become a substitute for the difficult, reciprocal work of relationships with other people. Both possibilities deserve testing rather than assumption.
Our assessment is that product success should not be measured only by longer sessions or more messages. For systems marketed around relationships, research access, clear boundaries and evidence about outcomes matter too.
The new loneliness report is an early warning worth investigating. It is also a reminder that a technology can become emotionally important long before the science has settled what that attachment means.
Featured image: Representative smartphone-use photograph; it does not show a study participant or establish anything about the user’s wellbeing. Photo: Jonas Leupe / Unsplash.
Leave a Reply