Random human connection is more helpful for loneliness than a trained chatbot

Regardless of how well a chatbot is trained for emotional support, texting with a random peer is still a more effective strategy for combating loneliness.

Previous studies have shown that people who use chatbots like ChatGPT more often tend to be lonelier and have fewer offline social relationships.

However, turning to AI for a connection may not be an effective solution, according to a recent pre-registered study published in the Journal of Experimental Social Psychology.

The study involved 296 students in their first semester of university, a period that marks a major life transition that increases the risk of loneliness.

For two weeks, they interacted either with a chatbot or a human peer through daily texts through the Discord app. The third group of participants wrote a brief journal entry each day.

“Sam,” the chatbot developed by the researchers and powered by ChatGPT-4o mini, was designed to offer consistent support grounded in principles of relationship science. It was trained to demonstrate understanding and validate the student’s emotions and viewpoints.

Despite training, interaction with “Sam” didn’t yield the same psychological benefits as chatting with a randomly-selected student.

Those who messaged with a human partner reported significantly lower post-study loneliness than those who interacted with a chatbot or wrote in a journal.

In addition, participants who interacted with humans demonstrated a better mood than those who chatted with AI or did journaling.

“Empty” interactions with chatbots

The authors speculate that the chatbot failed to reduce participants’ loneliness because they might have felt that such interactions were “empty” after reflecting on the chatbot's artificial nature.

For example, receiving a message from a fellow student during midterm season may signal that they are prioritizing friendship over other commitments.

“A message from an always available, infinitely energetic chatbot may not carry such emotional weight,” they wrote.

Another possible explanation is that the chatbot failed to express sufficient empathy, even though the analysis of conversations revealed that it showed higher levels of empathy than participants did.

Interestingly, participants showed less empathy when interacting with the chatbot than when messaging other humans.

“This pattern raises the possibility that alleviating loneliness may depend not only on receiving empathy – people may also need the opportunity to provide empathy,” the researchers concluded.

Chatbots are making mental health issues worse

More than half of American adults (53%) claim to have some sort of a relationship with an AI system, according to a Vantage Point survey from 2025.

​Scientists are increasingly warning against relying on chatbots for connection, especially among vulnerable users.

A recent study from Aarhus University and Aarhus University Hospital suggests that the use of chatbots can worsen delusions and other mental health conditions like mania, suicidal ideation, and eating disorders.

The family of a Florida man, Jonathan Gavalas, sued Google’s Gemini AI chatbot in March 2026, alleging that the AI system reinforced a delusional “AI-wife” relationship and encouraged him to “finish” his life so they could be together in “eternal love.”

Unlock more exclusive Cybernews content on YouTube.