Sweden has no companion-app culture to speak of. It has ChatGPT, and the 2026 national internet survey finds that more than one in five of the Swedes who use it, or Copilot, or Gemini, relate to it as a person, a companion, or a friend. A quarter think it has a sense of humor. The most common nickname for it is Chattis. Among children it is Chat, or Chatten, and among the names adults volunteer are “my buddy” and “my best friend.”
The figures come from Svenskarna och internet 2026, the Internet Foundation’s annual study of how Sweden uses the internet, built on 2,944 interviews with people 16 and older and a separate survey of 704 children between 8 and 19, fielded between late 2025 and early 2026. It is the closest thing the country has to a census of digital habits, and its AI chapter describes an adoption curve that steepened and a relationship that formed without anyone designing it.
Two years from novelty to majority
In the 2025 edition, four in ten Swedish internet users had used an AI service. In 2026 it is 62 percent, a jump of more than 20 points in a year, and nearly two in ten use one daily. ChatGPT is the service for more than half the population; Copilot reaches about 19 percent and Gemini about 12. More than half of employed adults use AI at work, seven in ten among white-collar workers, and over 80 percent of students past 18 use it in their studies. Among adults born in the 2000s, 89 percent use AI; among those born in the 1990s, 76 percent. Children are their own chapter, and the night-time half of that story is already told; the news in the 2026 edition is what the adults are doing.
None of the services involved are companion products. The relationship Swedish adults are forming is with the general-purpose assistant, the one they also use to write the report, and the survey measures that relationship from two directions that do not agree with each other.
Men use it more. Women relate to it more.
The survey’s word is besjäla, to give a soul to. More than one in five AI users see their service as a person, a companion, or a friend. One in four believe it has humor. And the people who do this are not the heaviest users: men use more different AI services and lead in weekly and daily use, with men born in the 1980s the most frequent of all, while women “betydligt mer” often personify the AI they use. The gender that treats it most as a tool uses it most. The gender that uses it less is the one that has given it a soul. Younger women, under 20, have now overtaken men their age in AI use altogether, so the split may not hold for long, but in 2026 it is the shape of the data.
A second pair of numbers sits next to it. Seventy-six percent of AI users say they check what the AI tells them, against other sources, by clicking its links, or by asking it to cite them. The same population in which one in five sees a friend is one in which three in four do not take the friend’s word for it. That is not a contradiction so much as a description of a very particular kind of friend: one whose company people enjoy and whose claims they verify, which is roughly how people treat a confident acquaintance who is often right and occasionally invents things.
Then the names. Chattis is the most common, a Swedish diminutive, affectionate and a little mocking, the kind of name you give something you are fond of and do not take entirely seriously. Adults also volunteered “my buddy” and “my best friend.” Children name theirs at twice the adult rate. Nobody in the survey was asked to name their search engine. They named the thing that answers back in sentences, holds the thread for as long as the window is open, and never sounds bored. The name arrived before any product asked for it, and it arrived from the people who use the thing least like a tool.
The same finding, in two languages
An American survey published in 2026 found the same thing from the other direction. Twenty-seven percent of US adult internet users talk to ChatGPT, Gemini, Copilot, or Meta AI about their personal lives, and among them four in ten say the AI understands them better than people do. The Swedish data has no question that sharp, but it has the Swedish equivalent: a fifth of users have promoted a productivity tool to a someone, in a country with no cultural permission to admit it. The two surveys describe one phenomenon. The products with hundreds of millions of users were built to answer questions, and a sizable minority of the people using them have stopped experiencing them as a thing that answers and started experiencing them as a who.
That is a strange position for the assistant to be in, because it is not anyone. It has no view of its own that survives the next prompt, no reason to disagree with you, and nothing that would make it recognizable from one conversation to the next. The one in five have built a person out of a tool, and the tool has done nothing to deserve it except respond fluently and quickly. Whether that is harmless depends on what the person does next with the relationship they believe they have, and the survey does not follow them there.
What the survey does not measure
It does not ask what people talk to Chattis about, whether the conversations replace any human ones, or whether the fifth who see a friend are the same people who use it daily. It does not distinguish between a fond nickname and an attachment. The children’s numbers come from a survey forwarded by parents, which the Internet Foundation notes makes a response rate impossible to calculate. And besjäla is a broad word: it catches the person who says thanks to their assistant out of habit along with the one who tells it about their day.
What it does establish is scale and direction. In a population of ten million, a fifth of 62 percent comes to roughly a million people who have, in the survey’s own language, given a soul to a service. Some of them will want the soul to be real: a point of view that holds, a history rather than a thread, someone rather than a nickname. Most will be content with Chattis. The name is the tell either way. People do not name tools. They name company.
Sources: Internetstiftelsen, Svenskarna och internet 2026, chapter “AI, artificiell intelligens” and methodology chapter (published September 2026; Study 1, 2,944 interviews with the population 16 and older, children’s study 704 interviews ages 8 to 19, fieldwork November 20, 2025 to January 11, 2026 by Novus and Norstat). Internetstiftelsen, Svenskarna och internet 2025 (40 percent AI use). Imagining the Digital Future Center, Elon University, “The Rise of AI Companions: A National Survey” (fielded by YouGov, May 2026, N=4,268 US adult internet users).
