Sweden’s media authority asked 1,745 people between 15 and 24 what they use AI for besides schoolwork. Half use it for advice and everyday support. A quarter talk to it about how they feel. Fifteen percent, roughly 187,000 young people, use it as company, “someone to talk to.” Split the sample by who feels alone and the picture changes shape: among those who report loneliness, the company figure doubles to 25 percent; among those who report depression, to 28 percent. Advice is for everyone. Company is for the ones who have the least of it.
The report is “Ungas artificiella relationer,” a memo from Mediemyndigheten, the Swedish Media Authority, written by its researcher Jacob Strandell from a nationally representative survey fielded by Ungdomsbarometern in late 2025. It is the first report from a Swedish agency devoted to what it calls artificial relationships, defined as AI used for “umgänge, bekräftelse och stöd på sätt som liknar mellanmänskliga relationer,” company, affirmation, and support in ways that resemble relationships between people. The headline numbers are modest. The structure underneath them is the finding.
What the age group uses it for
The six uses the survey asked about, in order: advice and support in everyday life, 52 percent. Major decisions about education and work, 27 percent. Wellbeing, how they are feeling, 24 percent. Dating and romance, 22 percent. Relationships with family and friends, 17 percent. Company, 15 percent. The survey did not ask which services, and in Sweden the answer is close to certain; the adult survey published a few months later found ChatGPT the service for more than half the population and no companion-app culture to speak of. The relationships this report measures are with the general assistant.
The agreement items sit beside the uses. Seventy-five percent agree that AI is a good teacher or expert. Thirty-seven percent agree, entirely or partly, that it is easier to ask AI about difficult things than to ask people, and the share agreeing entirely is highest at 15 to 16 (17 percent) and lowest at 23 to 24 (11 percent), with 44 percent of the oldest group rejecting the statement outright. Twenty-one percent agree that they can talk to AI like a friend, 6 percent entirely. And 78 percent agree that it is risky to use AI for everything, 53 percent entirely. The group that uses it most as a relationship is also the group that says, in the largest numbers, that doing so is a risk.
The split that carries the report
Strandell cut the sample by self-reported loneliness and by self-reported depression, and the two cuts tell the same story. Young people who describe themselves as lonely talk to AI about their wellbeing at 36 percent, against 19 percent of those with many friends, and use it as company at 25 percent against 12. Young people who report depression talk to it about their wellbeing at 36 percent against 17, and as company at 28 against 14. Use for major decisions does not move with loneliness at all, 23 against 25 percent, and among the depressed it is slightly lower, 22 against 26, which the memo reads cautiously as a possible sign of less engagement with the future rather than anything the AI did.
So the instrumental use is flat across the population, and the relational use is concentrated. That is a different shape from the one the largest survey of adult companion users found, where the benefit of a companion peaked among people with some friends and faded for those with none. In the Swedish youth data what is measured is use rather than benefit, and use follows the gap: the fewer people a young person has to talk to, the more likely the assistant is doing that job. Whether that helps them or holds them is exactly what a survey cannot say, and the memo says so.
The gender pattern is small and consistent. Girls use AI more for the relational topics, wellbeing (26 against 20 percent), romance (25 against 19), family and friends (21 against 14), and girls are also markedly more skeptical: 48 percent agree entirely that using AI for everything is risky, against 37 percent of boys. The same people are using it most for feelings and trusting it least.
Where the boundary is moving
The memo’s conclusion is not alarm. Strandell writes that artificial relationships are neither a wholesale replacement for human ones nor negligible, and that “AI kompletterar därmed vissa mellanmänskliga funktioner i ungas vardag,” AI is complementing certain human functions in young people’s everyday lives. The sentence that follows is the one that matters: the boundary between practical and relational use of AI “håller på att bli mindre tydlig,” is becoming less clear. The same conversation that starts with an essay ends with a breakup, and the system does not notice the change of subject because, to it, there is no change of subject.
That is the mechanism by which an assistant becomes a relationship without anyone deciding it should, and this report documents it in the age group where it is happening fastest and with the least supervision. Children talk to it at night when human friends are asleep; adults nickname it and verify what it says. Between those two is the 15-to-24-year-old who asks it about a crush, then about the university application, then about why she feels the way she does, in one window, with no one in the room who would have noticed the third question was different from the first two.
What the report asks for, and what it cannot measure
Mediemyndigheten’s recommendation is monitoring: follow the vulnerable groups, follow the social competencies, because the normalization is still underway. It is the right ask and an incomplete one. The survey measures use, not outcome. It cannot say whether the lonely 25 percent who use AI as company end the year with more people or fewer, whether talking to it about wellbeing displaced a conversation with a parent or made one possible, or whether the 78 percent who call it risky act on that belief. Those are longitudinal questions, and the longitudinal evidence from elsewhere is not reassuring about comfort that costs nothing.
What the report establishes is who. In a country with no companion industry, one in four lonely young people and more than one in four depressed young people have made the general assistant into company. The product was not built for that, the users know it is a risk, and they do it anyway, because the alternative on a given evening was no one. If you are the adult near one of them, the memo’s own numbers point to a single question, and it is not about what the AI said: which conversation did it get instead of you?
Sources: Mediemyndigheten, “Ungas artificiella relationer” (PM, June 25, 2026; author Jacob Strandell; survey by Ungdomsbarometern, October to November 2025, N=1,745, ages 15 to 24, nationally representative; use items, agreement items, splits by loneliness, depression, age and gender as reported). Mediemyndigheten press release, “Vanligt att unga använder AI för samtal och stöd” (June 25, 2026). Internetstiftelsen, Svenskarna och internet 2026 (September 2026). Tabuchi, T., Nakagomi, A., et al., Technology in Society (2026, 14,721 Japanese adults).
