Does sustained engagement with AI companions harm well-being?

The paper surveyed Character.AI users twice, 1,182 at baseline and 439 after a mean follow-up of 12 months. Interaction intensity, companionship use, and self-disclosure "all showed substantial continuity over time," and greater baseline interaction intensity predicted greater later intensity, companionship use, and self-disclosure. Sustained social engagement across these dimensions was "consistently associated with lower well-being," and the abstract reports that these links "were mainly explained by lower in-person social interaction."

The authors frame this as two hypothesized pathways. The longitudinal engagement pathway asks whether staying engaged over time relates to well-being. The social displacement pathway asks whether that relation runs through reduced human contact. The excerpt reports support for both, with displacement doing most of the explanatory work, so the engagement itself is not the main story. The discussion names the trajectory "from bowling alone to chatting alone." It adds a point about reciprocity. Unlike technologies that mediate interaction between people, an AI companion "can occupy users' social attention without reciprocal human participation." Family, partners, and friends lose interaction with the user too, which the authors call a possible second-order effect.

The introduction sets this against a split literature. Some studies find AI companions reduce loneliness and help users feel heard, while others link use to loneliness, problematic reliance, and reduced offline engagement. The paper argues the useful question is how engagement comes to shape well-being, not whether companions are good or bad. That sits alongside Do AI companions actually reduce loneliness like real people do?. The two findings differ in outcome (loneliness versus well-being), time scale (a week-long design versus about 12 months), and population, so this paper does not contradict the earlier one. It shows that short-term relief and long-run displacement can both be live questions. It also extends the "crowding out human partners" dynamic predicted in Do humans learn to prefer AI partners over time? with a year of field data on real users. Its finding of continuity, not decay, is a contrast worth holding against Do chatbot relationships lose their appeal as novelty wears off?, though the excerpt does not say the measures are comparable. It also bears on How should chatbot design vary by relationship duration?, since the persistent-companion pattern is the one this sample appears to occupy.

The excerpt does not establish causation. It says "predicts," "associated," and engagement "may influence" well-being, and it does not describe the design controls, the well-being measure, or effect sizes. It does not say how the 439 who returned differ from the 1,182 who started. The sample is Character.AI users only. The claim about reduced interaction for the people around users is an argument in the discussion, not a measured result. What follows at this strength is narrower than "companions harm well-being." The displacement of in-person contact is the quantity to watch when judging sustained use, and any account of companion benefit needs to say whether it survives that displacement.