Loneliness was once perceived as a personal issue, but we as a society are moving toward a more systemic stance. Governments and public health organizations have even labeled it an epidemic, linked to everything from depression to early mortality.^{2, 3} 

At the same time, AI systems are becoming more conversational, more embedded in daily life, and even, superficially empathetic. The solution, then, may seem obvious: AI is a scalable tool that can be designed to connect with people at times of loneliness. Unlike humans, chatbots don’t get tired or judgmental—they can be engineered to make people feel validated and seen. And in the short term, they seem to work.⁴ 

Previous studies show that people often feel better immediately after interacting with chatbots, even reporting similar boosts in mood and connection as they would after talking to another person.⁵ Some evidence even suggests that a week of chatbot interaction can reduce loneliness.⁶ 

Despite such promising findings, most of that research captures fleeting moments, telling us how people feel right after a conversation, or a few days later. Not much research explores what happens when that interaction becomes part of daily life, spanning multiple weeks, even years. That’s exactly what Li and colleagues set out to examine. 

The researchers designed a two-week experiment involving 296 first-year university students–a group particularly vulnerable to loneliness as they transition into a new social environment.

Participants were randomly assigned to one of three groups. In the AI condition, participants had daily text-based conversations with a custom-built chatbot called “Sam” that was designed to behave like an ideal supportive friend, using principles from relationship science. In the human condition, participants were randomly paired with another student and asked to text each other daily. In the control condition, participants wrote a brief one-sentence journal entry each day.

All interactions took place over private Discord chatrooms. Participants also completed daily surveys, as well as pre- and post-study assessments of loneliness using the UCLA Loneliness Scale.

While participants in both the AI and human conditions were instructed to send at least one meaningful message per day, they exchanged far more in practice, averaging around 8-10 messages daily.

So, have AI chatbots surpassed us in companionship? Rest assured, the results indicate that only human connection reduces loneliness. Participants who texted with another human showed a significant decrease in loneliness over the two-week period, while those who interacted with the chatbot did not.

Students in the AI condition reported loneliness levels that were statistically indistinguishable from those who simply journaled. Ultimately, the chatbot that was designed to be consistently supportive—arguably more so than what a random peer can provide—couldn’t produce the same psychological benefit.

Interestingly, when participants reflected on their individual interactions, there were no significant differences in how socially connected they felt in the moment across conditions. Chatting with AI felt just as meaningful as chatting with a person (at least temporarily). But over time, the findings demonstrate that feeling just didn’t translate into reduced loneliness.

However, the research doesn’t fully dismiss the impact AI chatbots had. The researchers also found that both AI and human interactions reduced negative moods compared to journaling. The bottom line? AI did have the ability to provide superficial emotional relief—it just failed to address the deeper, more persistent sense of loneliness.

In an analysis of participants’ behavior post-study, participants were more likely to continue chatting if they had been paired with a human rather than a chatbot. Some even exchanged contact information, suggesting the beginning of real relationships.

Exploratory analyses also revealed another intriguing pattern. While the chatbot expressed high levels of empathy, participants themselves were more engaged and more empathetic when interacting with another human. This suggests that deep connection may depend not just on receiving care, but on giving it.

An important caveat to point out is that the study lasted only two weeks and focused on first-year university students who generally reported relatively low baseline loneliness. Therefore, it’s unclear whether the results would differ over longer periods or in more vulnerable populations. That said, the findings add nuance to the discussion of therapeutic AI usage.

Li and colleagues challenge the growing assumption that simulating connection is enough to replace it. While AI companions may be excellent at delivering support, human relationships involve deeper connections developed through reciprocity and shared experience. Even though a chatbot is always available, its constant attention may not signal care in the same way a human does.

Given this, AI may be a useful tool for short-term emotional regulation, reducing minor stresses and negative moods, but it falls short as a substitute for meaningful social connection. The more promising direction may not be replacing human interaction but using AI to help people build and maintain real relationships.

Maybe after all, the best solution is often the simplest one. This study points to the value of basic, human-centered solutions. Pairing people for brief, repeated check-ins, whether in universities, workplaces or any communities, may be a low-cost way to reduce loneliness without having to rely on sophisticated technology.

If chatbots can’t replace the power of human connection, they may still help strengthen it. Rather than positioning AI as a companion, a potential path could be to design it as the first step in forming connections, where it functions to nudge people to reach out and follow-up with others. Using technology to assist in human connection formation could be a way to integrate the powerful technology we have around us for our own benefit. So, moving forward, the goal shouldn’t be to replace human connection but make it more likely to happen.

Across two weeks of daily interaction, only one thing consistently reduced loneliness: another human. Despite being designed to offer support, the AI chatbot did not reduce loneliness, performing no better than journaling. In contrast, even a randomly assigned peer led to meaningful improvements. The difference wasn’t just the change in medium but in how participants qualitatively felt during these interactions. In short, AI may help us feel better in the moment but it doesn’t necessarily make us feel less alone.

This article summarizes:

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, 104911. https://doi.org/10.1016/j.jesp.2026.104911

1. American Psychological Association. (2025, November 6). Poll reveals a nation suffering from stress of societal division, loneliness. https://www.apa.org/news/press/releases/2025/11/nation-suffering-division-loneliness
2. Ross, E. M. (2024, October 25). What is causing our epidemic of loneliness and how can we fix it? Harvard Graduate School of Education. https://www.gse.harvard.edu/ideas/usable-knowledge/24/10/what-causing-our-epidemic-loneliness-and-how-can-we-fix-it
3. American University College of Arts & Sciences. (2023, November 29). Surgeon General raises alarm: Is loneliness a national crisis? https://www.american.edu/cas/news/surgeon-general-raises-alarm-is-loneliness-a-national-crisis.cfm
4. Drouin, M., Sprecher, S., Nicola, R., & Perkins, T. (2022). Is chatting with a sophisticated chatbot as good as chatting online or face-to-face with a stranger? Computers in Human Behavior, 128, 107100. https://doi.org/10.1016/j.chb.2021.107100
5. Folk, D., Yu, S., & Dunn, E. (2024). Can chatbots ever provide more social connection than humans? Collabra: Psychology, 10(1), Article 117083. https://doi.org/10.1525/collabra.117083
6. De Freitas, J., Oğuz-Uğuralp, Z., Uğuralp, A. K., & Puntoni, S. (2025). AI companions reduce loneliness. Journal of Consumer Research, 52(6), 1126–1150. https://doi.org/10.1093/jcr/ucaf040