Mental Health Impacts of AI Companions: Triangulating Social Media Quasi-Experiments, User Perspectives, and Relational Theory

A team at Aalto University chose a method that sidesteps the weakness of surveys: it analysed the public Reddit posts of nearly two thousand active Replika users — a year before and a year after their first mention of an AI companion — and compared them with control groups. Plus 18 in-depth interviews.

Relief at first: posts about loneliness and crisis thin out. Then, as posts increasingly revolve around the relationship with the bot, they carry more signals of loneliness, depression and even suicidal thoughts than the comparison groups — and a pullback from human relationships. The interviews show why: a machine’s unconditional support quietly raises the price of human ties, which are unpredictable and take effort.

Presented at CHI 2026, the leading conference on human–computer interaction. For personal use this is a picture of a trajectory, not a snapshot: what helps in month one may harm in month twelve. Caveat: a quasi-experiment on public data; Reddit users are not representative, and the language of posts is an indicator, not a diagnosis.