A text message to a stranger may matter more than a conversation with artificial intelligence when it comes to combating loneliness, according to a study by researchers at the University of British Columbia.
Photo source: Pixabay
The study analyzed the behavior of 300 students in their first semester of college, a time when many young people feel alone and try to find their place in a new environment.
The participants were divided into three groups: some sent daily messages to another randomly chosen student, others wrote in a journal every day, and a third group was given access to a Discord server, where a chatbot based on the GPT-4o model was programmed to “listen actively and show empathy”.
After two weeks, students who communicated with a real person reported feeling about 9 percent less lonely. Those who spoke to the chatbot reached the same result as those who kept a diary – a difference of only 2%. The number of messages exchanged daily was almost identical in both situations, around eight to ten per day. However, the effects on loneliness were different.
“It's a very simple intervention that doesn't involve sophisticated technology, but it can make people feel a lot less alone,” said Ruo-Ning Li, a PhD student at the University of British Columbia, quoted by 404 Media.
The research specifically looked at the transition to college, a vulnerable stage when young people are away from home for the first time and are building new social networks. The question the researchers started with was whether a large language model could become a scalable tool for combating isolation. The answer, at least for this group, suggests no. Or not to the same degree as a real human connection, even with a stranger.
The study is part of a larger body of research looking at the long-term effects of chatbot use. For example, another study by the same researchers, published this week in the journal Psychological Sciencefollowed more than 2,000 people for a year, interviewing them every three months.
The researchers observed that the participants who spoke more with the chatbots felt more lonely in the immediate period. At the same time, it was precisely the students who felt the most alone who were also the most tempted to return to such conversations.
“Changes in the way people use chatbots seem to have only a small effect on feelings of loneliness. At the same time, the level of loneliness equally influences how likely a person is to turn to such conversations again.” said Dr. Dunigan Folk, one of the authors of the study, for 404 Media.
Chatbots, a kind of “social junk food”
But the study's authors say this doesn't necessarily mean chatbots cause loneliness. Rather, it is possible that people who already feel isolated seek out such interactions more often.
Why we reject compliments and how we can learn to receive them naturally
Folk points out that chatbots are a kind of “social junk food”: they provide a pleasant momentary state, but they cannot replace real relationships between people.
“They may make you feel good for a moment, but in the long run they may not provide the same kind of support that human relationships do.” he says.
He believes that for some people, conversations with AI models end up taking the place of interactions with other people. “I think it's ultimately a trade-off: you're talking to the AI instead of a person”he says. “And talking to a person would have been much more satisfying.”
But there are also indications that interacting with artificial intelligence can have short-term positive effects on mood. More specifically, Ruo-Ning Li admits that people feel better after a conversation with a chatbot. “But making someone feel good for a few minutes is no big deal.” she thinks.
It is not clear, however, whether a momentary positive experience can have the same long-term effect.
“It's like when you eat candy and you feel good. But if you eat a lot of candy for a long time, things can get unhealthy.” Li said.
What happens when you try a chatbot as a therapist
A similar but more personal experiment was published in early March in The Guardian. Journalist Rhik Samadder, a self-proclaimed AI skeptic, used ChatGPT as a therapist for several weeks and says he was surprised by the results.
Samadder had written to the chatbot about the exhaustion of being the sole caregiver for his 82-year-old mother. The bot's response came with a concrete plan for organizing tasks and rewordings aimed at reducing emotional tension. Halfway through the answer, he admits, he started to cry.
He claims, however, that the experience left him with a conflicted feeling. The chatbot gave him clarity and practical steps he could follow, but not humor, not a relationship, not wisdom accumulated over time. The difference from real therapy, he says, comes down to something harder to define: “a relationship built over time, with a man who judges you, sometimes contradicts you and stays in your mind long after the session is over”. The therapist's voice, he claims, still echoes in his mind today – he internalized it. A chatbot, however useful it may be in the moment, does not leave traces of the same kind.
Moreover, he points out, there are moments in a man's life (overwhelming news, various forms of loneliness) that should not be handled in four seconds in front of a screen, by software that has neither thoughts nor responsibility to the one in front of it.
A study conducted in March 2025 by the MIT Media Lab together with OpenAI reached similar conclusions. Over the course of four weeks, researchers tracked how different types of chatbot conversations affect users' well-being. Although some interactions seemed beneficial at first, those who used chatbots daily and heavily reported a greater sense of loneliness, a tendency toward addiction, and less time spent with real people.
