New Character.AI study published on arXiv
arXiv
Character.AI users who chat more keep chatting more, a new study says. That long-term habit tracks with lower well-being, mostly because people see fewer humans in person.
Based on reporting by arXiv — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
A new longitudinal study of Character.AI users says the habit of talking to AI companions doesn’t just fade away. It tends to stick. Researchers surveyed 1,182 people at the start and followed up with 439 of them after a mean of 12 months, then looked at how intensity of use, companionship, and self-disclosure changed over time.
The pattern was pretty clear: people who were more active at baseline were also more active later. Higher early interaction intensity predicted more later interaction, more use for companionship, and more self-disclosure. In other words, the people leaning hardest on the system at the start were also the ones most likely to keep leaning on it.
That matters because the study found a downside. Across those engagement measures, sustained social use of the chatbot was linked with lower well-being. The researchers say that connection held up over time, not just in a one-off snapshot.
And the main explanation wasn’t mysterious. The link between heavier AI companionship and worse well-being was largely driven by lower in-person social interaction. So the concern is not simply that chatbots are emotionally sticky. It’s that they may crowd out the human relationships that actually do the work.
The authors say this points to a design problem for AI companions: make them supportive without turning them into a substitute for real-world social life. That is a neat line on paper. It is much harder product work in practice.
My take — AI-written commentary, not fact-checked reporting
This is the part the AI booster crowd keeps waving away: “companionship” sounds harmless until it starts replacing the boring, necessary business of seeing actual people. A chatbot that helps someone through a rough patch is one thing; a product that quietly trains users out of human contact is another. The industry keeps selling emotional features like candy, then acting shocked when the nutrition label says sugar.
Read more about this at: arXiv
