Gemini improves tutoring and classroom collaboration
Two Google studies in the United Kingdom and Italy show how Gemini can support math tutoring, increase classroom collaboration, and reduce some of the pressure on teachers. The results also reveal limits: improvements in tasks and quizzes have not yet appeared in standardized tests.
Google says Gemini can help students learn math and change classroom dynamics, but the results depend on how it is used. Two studies in the United Kingdom and Italy show gains in tutoring, participation, and teacher well-being, along with an important warning: better day-to-day understanding does not always translate immediately into better standardized test results.
United Kingdom: more personalized math tutoring
Over 10 weeks, more than 1,100 students aged 12 to 15 took part in a controlled trial at British schools. The study was conducted on the Eedi education platform and compared three types of support when a student answered a question incorrectly:
- A fixed hint.
- Help from a human tutor.
- Interactive tutoring powered by Gemini and Eedi's teaching resources.
Teachers closely tracked students' progress and provided support in the classroom. Human experts also reviewed every AI-generated response before it reached the student. It was not, therefore, a fully autonomous system.
According to the results published by Google, the model's pedagogical quality improved by almost 25% compared with the previous year's study. The share of responses directly approved by reviewers rose from 76% to 95%.
In platform tasks and interim quizzes, students who received help from Gemini performed similarly to those who worked with human tutors. The AI also avoided simply giving them the answer: in almost 99% of sessions, it guided students to reason through the problem, and students solved the math themselves in more than 85% of conversations.
There is, however, a clear limitation. The improvement seen in tasks and quizzes did not appear in STAR standardized tests during those 10 weeks.
The proposed explanation is practical: step-by-step tutoring takes more time than a brief hint. Students solved an average of 14 problems per session with AI, compared with 19 with static hints. By covering fewer exercises, they also progressed through less of the curriculum, even though their understanding and reasoning appeared to improve.
Google is preparing studies lasting the entire school year to find out whether that understanding eventually shows up in broader assessments. For now, the conclusion is not that AI replaces the teacher, but that it can provide individualized support when there is supervision and enough time.
Italy: less lecturing and more group work
The second study took place over a year at 31 Italian schools, in collaboration with the Salesians of Don Bosco network and the Pontifical Salesian University in Rome. It involved 1,600 teachers, representing more than 24,500 students aged 13 to 18.
The team compared classroom activities and lesson plans created with and without Gemini for Education. The goal was to observe not only learning, but also student participation and teacher well-being.
Teachers who used Gemini to design their lessons said they felt more creative in their planning: 79%, compared with 30% when they did not use the tool. Lecture-based teaching fell by 38%, while student-led collaborative activities nearly tripled.
Participation also increased. The share of students reporting that they were engaged rose from 77% to 84% in classes their teachers had prepared with Gemini. In those classes, students discussed more viewpoints and took greater responsibility for their learning.
The study also found an effect on teachers' emotional workload. Reliance on the teacher's personal energy to keep the class engaged fell by 66%. At the same time, the share of classes in which the activity design itself generated participation rose from 23.2% to 52.2%.
For you, the distinction matters. Gemini does not provide the same value when it is limited to generating answers as it does when used as a planning partner, a tutor with clear boundaries, or support for adapting materials. In both studies, the teacher's role remained central.
What remains to be monitored is whether these benefits last over time and lead to broader academic results. The United Kingdom experience is a reminder that AI can help students think better without automatically increasing the amount of content they learn. In the classroom, the decisive question is not only what Gemini can answer, but how it fits into teaching guided by people.