Google announced the results of new research showing how the Gemini artificial intelligence model contributes to supporting students and teachers across classrooms in both the United Kingdom and Italy. The studies focused on evaluating the effectiveness of AI tools in providing personalized support for mathematics and assisting teachers in planning interactive lessons that enhance classroom engagement.
In the United Kingdom, a 10-week randomized controlled trial was conducted involving more than 1,100 students aged 12 to 15 in collaboration with the learning platform Eedi. The experiment utilized a system combining Gemini's capabilities with the platform's resources to guide students toward resolving errors themselves step by step. The results showed that the quality of Gemini's responses improved by about 25%, with the approval rate from expert reviewers rising to 95%, as students who received AI guidance achieved results comparable to those of students who relied on human teachers.
In Italy, a year-long study was conducted in cooperation with the Don Bosco school network and the Pontifical Salesian University in Rome, involving about 1,600 teachers and more than 24,500 students. Gemini was used as a lesson design assistant, which led to a 38% decrease in traditional direct lectures in favor of workshops and group activities, an increase in active student participation from 77% to 84%, alongside a reduction in daily stress for teachers.
What does this development change practically?
These findings highlight a practical shift in how artificial intelligence is integrated within educational institutions, where its role is not limited to providing ready-made answers, but rather stimulates critical thinking among students (including guiding them without giving the direct solution in about 99% of sessions). It also contributes to reducing complete reliance on individual teacher energy to maintain classroom engagement, allowing more space to focus on personal mentoring and guidance.
Limitations and challenges associated with the experiment
Despite the clear improvement in student performance during daily tasks and short quizzes, these gains did not appear significantly on standardized STAR test scores during the same short timeframe of the experiment. According to researchers, this is because step-by-step guidance takes more time per question compared to simple visual hints, which reduced the number of problems students solved in a single session, something that will require study across a full academic year in the future.