HELSINKI, Finland — Finland, whose school system has been studied by visiting ministers for two decades, released the results on Tuesday of the largest controlled trial of A.I. tutoring software in Europe, and the headline finding was deliberately unglamorous: it helped some children a little, others not at all, and in one subject it may have done harm.
The two-year study, run by the National Agency for Education with researchers at the University of Jyväskylä, followed 21,000 students aged 11 to 15 in 180 comprehensive schools. Half of the classrooms used a tutoring assistant for math and Finnish-language writing; the other half were taught as usual.
In mathematics, students in the bottom third of their classes gained the equivalent of about six weeks of additional learning over two years, a modest but statistically solid effect. Students in the top third gained nothing measurable. In writing, the picture inverted: scores on supervised in-class essays fell slightly among frequent users, and teachers reported that some students could no longer begin a paragraph without the tool’s opening sentence.
“This is exactly the kind of result we wanted, because it is true,” said Antti Koskela, who directs digital learning at the agency. “A vendor would tell you the software transforms everything. Our data say it is a decent assistant for a child who is lost in fractions, and a crutch for a child who is learning to think on paper.”
The trial cost about 34 euros per student per year, including teacher training, which Mr. Koskela said was the item most schools skimped on. Classrooms whose teachers completed a 20-hour course saw roughly twice the math gains of those that did not.
Riikka Mäkinen, who teaches math at a school in Tampere, said the tool’s best feature was its patience. “A 13-year-old will ask a machine the same question five times. He will not ask me five times, because the class is watching,” she said. But she now bans it during the first draft of any written assignment. “The thinking has to happen before the typing.”
Laura Saarinen, the Jyväskylä researcher who led the analysis, cautioned against reading the writing result as a verdict on the technology. “It is a verdict on unsupervised use,” she said. The study will be published in a peer-reviewed journal next month; a second phase, limited to math and to schools with trained teachers, begins in January.
Eemeli, a 14-year-old in Ms. Mäkinen’s class who was identified only by his first name under the study’s rules, offered his own summary. “It is good for homework,” he said. “For the test, it is just me.”
A version of this article appears in print on Oct. 12, 2026, Section A, Page 8 of the European edition with the headline: Finland Put A.I. Tutors in 180 Schools. The Results Were Mixed, and Teachers Say That’s the Point..
