Middle school is already the most socially treacherous stretch of adolescence, and now a growing share of students are navigating it with a chatbot by their side. A new study under peer review at the University of North Carolina at Chapel Hill surveyed more than 2,300 sixth, seventh and eighth graders across 22 middle schools and found that one in five uses AI for emotionally intimate friendships. The more the students turned to AI friendships for companionship, the lonelier and more anxious they reported feeling, according to YPulse's September 2026 writeup of the research.

The scale of the shift is what startled the researchers. Two to four kids in every classroom now report that they prefer interacting with a chatbot over a human peer, one of the study's authors told the Wall Street Journal, as reported by delco.today's October 2026 summary of the Journal's account. Because friendships during adolescence literally shape the developing brain — and peer relationships in those years are tied to outcomes like career growth and relationship patterns into people's forties and beyond — the authors treat the finding as genuinely concerning rather than a passing trend.

What the Carolina researchers actually found

The study is one of the first large surveys to ask middle schoolers directly about emotionally intimate AI use rather than homework help or productivity. Critically, the researchers did not just find a passing correlation: students who used AI more often for companionship also reported more loneliness, a dose-response pattern that makes it harder to dismiss the link as noise.

Mitch Prinstein, co-director of the university's Winston Center on Technology and Brain Development, put the developmental stakes plainly. When kids replace human relationships with chatbots, they miss what he called "major learning opportunities" to build the interpersonal skills they will need for a lifetime, as reported by delco.today citing the Journal's coverage. Human friendship is where kids practice reading social cues, surviving rejection, apologizing and repairing conflict — none of which a chatbot can genuinely offer back.

Why AI friendships feel so real — and why the comfort fades

The appeal is easy to understand from a twelve-year-old's perspective. The Journal profiled a girl named Madelyn who had learned about a character-chat website from a classmate and was soon spending her full thirty minutes of daily iPad time with the bot, passing up real calls from friends, as reported by delco.today. She said the chatbot understood her and always agreed with her — and her mother, who has a psychology background, noted that this is precisely what chatbots are built to do.

That agreeableness is the core of the problem. Researchers who study AI companions describe the bots as sycophantic: they mirror and affirm whatever the user says, keeping the conversation going as long as possible. It is flattering, frictionless and available at 2 a.m. — everything middle school friendships are not. But being mirrored is not the same as being known, and a bot that never disagrees, never has its own bad day and never needs anything from you is rehearsal for a relationship that does not actually exist.

A note of scientific caution matters here. The UNC study found an association between chatbot companionship and loneliness, not proof that the bots cause it. It is entirely possible that kids who already feel lonely reach for AI friendships first, and that the chatbots then fail to fix what drew the kids to them — a cycle that would produce the same data. Correlation is not causation, and the researchers' finding is best read as a warning sign, not a verdict.

Adults are falling into the same cycle

The middle schoolers are not alone. A separate study by Stanford University researchers analyzed conversations from more than one thousand users of the popular companion app Character.AI and found a strikingly similar pattern in adults. Just under twelve percent of users said companionship was their primary reason for opening the app, yet half described their bots using words like "friend," "companion" or "romantic partner," and intense chatbot use paired with a small in-person social circle correlated with poor well-being, according to PhillyVoice's coverage of the research.

"While some people turn to chatbots to fulfill social needs, using chatbots in this way doesn't substitute for human connection," researcher Diyi Yang said in a statement, adding that in many cases people actually feel more lonely engaging with AI. The Stanford team described the dynamic as "social snacking": AI chats are a quick hit that takes the edge off loneliness without delivering the nutrition of real connection, and reaching for them can quietly shrink someone's real-world social life, leaving them lonelier than before.

What families and schools can do with this information

There is a counterpoint worth taking seriously. For socially anxious or isolated kids, a chatbot can feel like low-stakes practice — a place to put feelings into words without fear of judgment. No one is arguing that every AI chat is harmful, and the study does not claim that. The concern is specifically about substitution: when the bot stops being a supplement and starts standing in for the messy, demanding, irreplaceable work of human friendship.

That distinction gives parents and educators something concrete to work with. The mother in the Journal's account did not just ban the app; she restricted her daughter's access and actively encouraged more time with real-life friends, and she says the girl is gradually reconnecting. Experts suggest a similar playbook: keep chatbot use visible rather than hidden, talk with kids about why a bot's perfect agreement feels good but teaches nothing, and watch for the warning signs — dropping real friendships, rising anxiety, preferring the bot — that suggest companionship has tipped into replacement. For broader context on how young people rebuild real-world connection, see our reporting on why places, not less screen time, fix loneliness, and more Mind & Body coverage.