Google published results from two studies of its Gemini AI tools in classrooms in the United Kingdom and Italy. The company said AI tutoring gave U.K. students math help on par with human tutors on daily tasks, and that Italian teachers who used Gemini to plan lessons reported more collaborative classes.

The findings are Google’s own account of the research, which it said it carried out with educators and researchers in both countries.

In the U.K., Google worked with Eedi, a math platform used by schools. The company said it ran a 10-week randomized controlled trial, a study that assigns participants to different groups by chance, with more than 1,100 students ages 12 to 15. When a student answered a question incorrectly, the student received one of three types of help: a static hint, help from a human tutor, or interactive AI tutoring that combined Eedi’s teaching resources with Gemini.

Google said human experts reviewed every AI response in real time before it reached a student. Teachers monitored progress and gave hands-on support in the classroom. According to Google, expert reviewers fully approved 95% of Gemini’s responses, up from 76% in last year’s study.

Google said students who received AI tutoring performed similarly to those paired with human tutors on in-platform tasks and intermediate quizzes. The company said that, following a prompt it provided, Gemini guided students instead of handing over answers in nearly 99% of sessions. It said students worked through the math themselves in more than 85% of their chats.

The gains did not show up on STAR standardized test scores over the same period, Google said. The company said step-by-step tutoring takes more time per question, so students solved an average of about 14 problems per session with AI tutoring, compared with 19 with static hints. Google said it believes the 10-week trial was too short to cover the range of topics on the STAR test, and that it is exploring follow-up trials across a full academic year.

In Italy, Google worked with the Salesians of Don Bosco school network and the Pontifical Salesian University in Rome on a yearlong study across 31 schools. The study drew on 1,600 teachers, who together represent more than 24,500 students ages 13 to 18. Researchers collected classroom activities and teachers’ self-reports on lesson planning, both with and without Gemini for Education.

Google said 79% of teachers who used Gemini to co-design lessons reported feeling more creative than usual, compared with 30% without it. Traditional lectures at the front of the classroom dropped by 38%, the company said, while student-led collaborative learning nearly tripled. Student-reported engagement rose from 77% to 84% in lessons designed with Gemini.

Google also reported effects on teacher wellbeing. In lessons co-planned with Gemini, teachers’ reliance on their own energy to keep students engaged fell by 66%, the company said. Lessons in which the activity design itself drove engagement rose from 23.2% to 52.2%.

Google said full technical and implementation reports are available, including a report from the Don Bosco CNOS-FAP Foundation on the Italy study.