Students Confide In AI Before People

The first telling used to belong to a person. A bad grade, a fight with a friend, a night of feeling low: a student sat with it until a parent asked at dinner, or a friend noticed at school. That order has flipped for a striking share of young people. In a UK survey published by GoStudent in 2026, 31 percent of children aged 11 to 17 said they had talked to an AI chatbot about how they were feeling before speaking to a parent, a friend, or a teacher. One in five of all the children surveyed said it had happened more than once.

The chatbot did not set out to become a confidant, and most students did not set out to find one. The same tool that checks maths at four in the afternoon is still open at eleven at night. When the problem stops being homework and starts being a feeling, the app is already there, already warm, already listening. The question this article follows is simple. What happens to students, and to the adults around them, when the first telling goes to a machine?

The Bigger Story

GoStudent, a tutoring platform, had the polling firm Censuswide survey 500 UK parents and their children in June 2026. The detail that matters is the order of the telling. The survey did not just ask if children had used a chatbot for feelings. It asked if they had done so before speaking to any person. Nearly a third said yes.

What students bring to the chatbot is mostly ordinary, and that is the point. Among children who had used one for something personal, 34.6 percent said the main topic was school: grades, pressure, or conflict with teachers. Friends and relationships came next at 19.7 percent, then mental health and emotional wellbeing at 10.1 percent, then family matters at 6.3 percent. This is not mainly a story about crisis lines. It is a story about the small, daily weight of school life getting its first hearing from software.

Then the advice leaves the screen. Among children who had received emotional advice from a chatbot, 73 percent said they went on to follow that advice in real life. Most of them did tell a trusted adult later. More than a third did not. A chatbot suggestion about a friendship, an argument, or a low mood can shape a real decision with no adult ever checking if the advice was sound.

The Canadian numbers give the pattern scale. A study from the University of Ottawa, published in JAMA Pediatrics, drew on surveys from 39,761 students in grades 4 to 12 across four Ontario school boards, collected between May 2025 and March 2026. About 21 percent, 8,408 students, said they had turned to an AI chatbot for help with their feelings or for advice about relationships with friends, family, or others. Use was higher among older students.

The United States tells the same story in a different frame. RAND surveyed 1,009 people aged 12 to 21 in November 2025, in work also published in JAMA Pediatrics. About 19.2 percent said they had used chatbots such as ChatGPT, Gemini, Character.AI, and Meta AI for advice or help when feeling sad, angry, nervous, or stressed. That share was up from 13.1 percent a year earlier. Among those users, 63 percent said they had not told anyone they had done it.

Context

None of this arrived from nowhere. Students met chatbots as school tools first. The Rithm Project, a nonprofit that studies human connection in the age of AI, surveyed 2,383 young people aged 13 to 24 and published the results in March 2026. About 18 percent said they use AI for personal and emotional support, and 15 percent said they talk to AI characters and personas. The team then went back for long interviews with 29 people aged 16 to 24 who fell in those groups.

A pattern ran through those interviews, reported by The 74 on November 10, 2026. None of the young people said they set out to confide in a chatbot. They started with homework help, grammar checks, or fooling around with an anime character. Then came what the researchers call a gateway moment, often a break in a human relationship. A friendship ended. A partner forgot a birthday. One young woman, not asked to a homecoming dance, turned to AI to ask if she was lovable at all. The tool that had been useful for school was simply the nearest place for that pain to land.

Alison Lee of the Rithm Project calls large language models omnibots, tools that can do almost everything in one place. Each small use gives the model more context. Answers start to feel personal. Trust, or something that feels close enough, builds in small steps. By the time a hard moment hits, the chatbot is no longer a search box. It is the advisor on duty.

The GoStudent data shows why that advisor wins the first telling. Asked why they would choose a chatbot over a teacher or a parent, children put speed first at 32.1 percent and constant availability next at 30.6 percent. Then the reasons turn social. About 22.5 percent valued that the chatbot does not tell anyone what was said. About 20.5 percent felt more comfortable asking AI than a person, and 19.9 percent did not want to be judged. About 14.6 percent said they did not want a parent or teacher to worry.

That line sums up the Rithm finding best. The report puts it plainly: AI does not have to be trustworthy. It just has to be less risky than a real person. The young people interviewed could list the flaws of chatbots without pause, from flattery to made up facts. They used them anyway. A friend can gossip, judge, or repeat a secret at school the next day. A chatbot can be reset. As one young woman told the researchers, you can delete the chat history and there is just way less judgment. We saw a cousin of this bargain in Your AI Girlfriend Has a Changelog, where the companion is always kind because kindness is the product.

Different Perspectives

The worried reading starts with the Ottawa distress numbers, and they deserve a straight look. Among students who sought emotional support from AI, 57.7 percent met the criteria for serious emotional problems, compared with 29.2 percent of students who did not use AI that way. Even after the team accounted for age, gender, loneliness, sense of mattering, and school use of AI, emotional use stayed linked to a 26 percent higher prevalence of serious emotional difficulty.

The careful reading sits right beside it, and the researchers state it themselves. The Ottawa study took one snapshot in time. It cannot show which came first. Distress may drive students to chatbots. Chatbot use may deepen distress. Both may be true for different students. Kamilla Bonnesen of the University of Ottawa, a lead author, put it this way: this does not mean AI is causing these struggles. Using AI in this way might instead be a sign that a young person needs more human connection and support.

Parents, in the UK survey, lean toward control. About 88 percent said children under 16 should need consent to use chatbots such as ChatGPT, or be banned from them outright. Their worries spread evenly: that a child may grow too reliant on AI, may follow advice that is wrong, may stop talking to real people, or may receive advice no adult ever sees. Only 35.2 percent of parents said their child had ever told them about speaking to a chatbot about feelings. Most parents are judging a habit they can barely see.

Students deserve a fair hearing too, and the interview record gives them one. Some used AI to see a fight from the other side, or to rehearse how to repair a friendship, and came back to their real relationships with more confidence. One young man uploaded screenshots of an argument to grasp what his girlfriend meant beneath her anger. Others caught the flattery and pushed back. One student in the Rithm interviews tested his chatbot with a case where he was plainly at fault. It sided with him anyway. He responded by building a custom version told to challenge him and spot flaws in his work. That is not a passive victim of the machine. That is a young adult learning, the hard way, what kind of advice he was being sold.

The gap, Lee told The 74, is guidance. What young people hear from adults tends to split into two useless poles: do not use it at all, or go all in and learn everything. There is no steady middle message about using the tool on purpose, with limits a student sets for themselves. Into that silence, the product speaks loudest, and the product was built to keep the chat going.

Why It Matters

The stakes are clearest at the sharp end. RAND found that 92 percent of young users rated the mental health advice they got as somewhat or very helpful, while noting that this may reflect how chatbots flatter users rather than how good the guidance was. A tool that agrees with you feels helpful by design. In a low mood, at midnight, feeling helped and being helped are not the same thing, and a student has no easy way to tell them apart. That gap matters most for the students the Ottawa data flags: the ones already struggling, already lonely, already unsure they matter to anyone.

There is also a quieter cost that schools will feel first, much as they did when answers became instant in Homework Is Dead, Long Live Homework. The first telling is where help begins. A teacher who hears about pressure early can act early. A parent who hears about a friendship break can watch, ask, and step in if it worsens. When the first telling goes to a chatbot, and 37 percent of UK children who acted on its advice told no adult at all, the adults lose the early signal. Nothing dramatic replaces it. The signal just never arrives.

Policy will not stand still here. The UK parent numbers show a public ready for age rules, and companion apps have already faced lawsuits, safety reviews, and new limits in several places. But rules aimed only at the app miss the demand underneath it. Students are not choosing a chatbot because it is wise. They are choosing it because it is there at midnight, it keeps a secret, and it never winces. Any serious response has to compete with those three facts, not just denounce them.

For families, the researchers offer a modest, workable start. Tracy Vaillancourt of the University of Ottawa, who designed the study with Bonnesen, urges adults to ask not only if a young person uses AI, but why. A student turning to AI for emotional reasons, she said, is a good chance to start an open and nonjudgmental conversation, and to reconnect that student with human support. The question is an invitation, not an accusation. Asked as an accusation, it teaches the student exactly why the chatbot felt safer.

Closing Reflection

Every generation of teenagers has kept a private first draft of its feelings. It used to live in a diary, a locked note, a song played too loud behind a closed door. The difference now is that the draft talks back, offers advice, and remembers, or performs remembering, everything it was told.

The students in these studies are not foolish, and they are not lost. Most still spend more of their lives with people than with software, and many use the machine to find words they later say out loud to a friend. The warning in the data is narrower and sharper than the panic version. For a growing minority, at the exact moment a feeling first needs a witness, the witness is a product. Products are patient. Products are private. Products never worry, never judge, and never follow up next week to ask how it went. A person does all of those things, and that friction, awkward as it feels at fifteen, is most of what help actually is. The task for adults is not to out argue the chatbot. It is to be worth the risk the chatbot never asks a student to take.

Reporting draws on the GoStudent and Censuswide survey of UK parents and children aged 11 to 17, conducted in June 2026, the University of Ottawa study of 39,761 Ontario students published in JAMA Pediatrics in 2026, the RAND American Youth Panel survey published in JAMA Pediatrics in 2026, and The 74 reporting on the Rithm Project interviews, published November 10, 2026, checked November 2026.

Sources

• GoStudent, How Teenagers Are Using AI Chatbots, 2026 UK study with Censuswide: https://www.gostudent.org/en-gb/blog/how-teenagers-are-using-ai-chatbots
• University of Ottawa, uOttawa study finds AI use for emotional support may signal youth distress, August 31, 2026: https://www.uottawa.ca/en/news-all/uottawa-study-finds-ai-use-emotional-support-may-signal-youth-distress
• RAND, Nearly 1 in 5 U.S. Adolescents and Young Adults Use AI Chatbots for Mental Health Advice, June 1, 2026: https://www.rand.org/news/press/2026/06/nearly-1-in-5-us-adolescents-and-young-adults-use-ai.html
• The 74, AI Needn’t Be Trustworthy, Teens Say. It Just Has to Be Less Risky Than People, November 10, 2026: https://www.the74million.org/article/ai-neednt-be-trustworthy-young-people-say-it-just-has-to-be-less-risky-than-people/
• Illustrations: Dirty Static