I write a lot about AI in classrooms, but the wider question of what AI does to student wellbeing is one I keep returning to. Loneliness is a real problem among first-year university students. AI companion apps are increasingly marketed to young people as a solution.
Li, Folk, Singh, Ungar, and Dunn (2026) published a pre-registered randomized controlled trial in the Journal of Experimental Social Psychology with an interesting finding: texting daily with a random peer reduces loneliness. Texting daily with a highly supportive chatbot does not.
AI Chatbots and Student Loneliness
The research team didn’t use an off-the-shelf chatbot. They built “Sam,” powered by ChatGPT-4o mini, and engineered the prompt around relationship-science principles: active listening, empathy, validation, and ongoing memory of the user. Sam was designed to embody the qualities of an ideal friend.
Then they recruited 296 first-year students at a Canadian university and randomly assigned them to one of three conditions for 14 days: text daily with Sam, text daily with a random fellow first-year student, or write a one-sentence daily journal as an active control. All interactions happened through Discord. The researchers measured loneliness with the UCLA Loneliness Scale and tracked mood, social connection, and engagement throughout.
What They Found
Only the human-peer condition showed a significant reduction in loneliness. The chatbot condition did not differ from the journaling control on loneliness.
The most striking finding came from the conversation analysis. The researchers used GPT-rated empathy scores (validated against human coders) across roughly 2,000 daily conversations. They found that Sam expressed substantially higher empathy than any human participant. The chatbot was, by their measurement, a better empathy giver than any of the students.
The twist is unexpected. Participants who texted with Sam expressed less empathy themselves than participants paired with a human peer. The receivers became less giving.
This led the authors to a useful conceptual move. As they put it, “alleviating loneliness may depend not only on receiving empathy; people may also need the opportunity to provide empathy” (p. 6).
That single sentence reframes what loneliness needs. The chatbot can’t be lonely. It doesn’t benefit from your care. And when there’s nothing for the user to give, the human in the conversation is reduced to a passive recipient of artificial warmth.
Connecting to Other Recent Work
The finding aligns with Claessens, Veitch, and Everett’s (2026) work on the negative perceptions of outsourcing to AI, and with Shaw and Nave’s (2026) cognitive surrender framework. Both papers point at the same underlying problem: when AI does too much of the human work, something on the human side of the equation atrophies. Li and colleagues add a relational version of that story. The capacity to care for another person is itself something that can atrophy when there’s no one on the other side to receive it.
The behavioral data backs the survey findings. After the study ended, participants could keep using their Discord rooms voluntarily. Only 3 percent of the journaling group continued, 14 percent kept talking to Sam, and 33 percent kept texting their human peer. Thirty-seven percent of human-condition participants also exchanged contact information. The chatbot didn’t generate a relationship that anyone wanted to continue.
My Take
I find this paper persuasive on the central claim. The pre-registration is solid, the chatbot was thoughtfully built, the comparison condition is exactly the right one, and the empathy-giving finding adds real conceptual value.
Where I’d push: the sample is undergraduates with relatively mild baseline loneliness. The paper itself flags this. People in deeper isolation, older adults, immigrants, or students recovering from grief may respond differently to AI companionship. Pew Research’s recent work on how teens use AI (McClain et al., 2026) shows that young people’s relationship with AI tools varies significantly. The story may look different for populations the current study didn’t sample.
The study also lasts only two weeks. The chatbot field is moving fast. A six-month or year-long version of this experiment, with a more capable model and more sophisticated relational memory, could yield different patterns. The authors are appropriately cautious about generalizing, and I’d echo that caution in both directions.
What this paper does change is the default assumption. Many AI companion products are marketed on the premise that a highly supportive chatbot beats no one at all. This study tells us that’s not quite right. For young people undergoing a life transition, the better intervention may be the simplest one: pair them with another person going through the same thing.
The authors close with a line that sticks. “Perhaps the true value of AI companions will not be in how closely they can mimic human connection, but how well they help us sustain connections with one another” (p. 5). That’s the design imperative worth taking seriously. AI works best as a bridge to other people, not as a replacement for them.
References
- Claessens, S., Veitch, P., & Everett, J. A. C. (2026). Negative perceptions of outsourcing to artificial intelligence. Computers in Human Behavior, 177, 108894. https://doi.org/10.1016/j.chb.2025.108894
- McClain, C., Anderson, M., Sidoti, O., & Bishop, W. (2026, February 24). How teens use and view AI. Pew Research Center. https://www.pewresearch.org/internet/2026/02/24/how-teens-use-and-view-ai/
- Li, R.-N., Folk, D., Singh, A., Ungar, L., & Dunn, E. (2026). Is a random human peer better than a highly supportive chatbot in reducing loneliness over time? Journal of Experimental Social Psychology, 125, Article 104911. https://doi.org/10.1016/j.jesp.2026.104911
- Shaw, S. D., & Nave, G. (2026). Thinking fast, slow, and artificial: How AI is reshaping human reasoning and the rise of cognitive surrender. Working paper, The Wharton School, University of Pennsylvania. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646