How Gemini is helping students and teachers in the classroom

What it claims

Google reports that an AI tutor built from Gemini and Eedi's pedagogical resources performed on par with human tutors on in-platform work for UK students aged 12 to 15. This held only with every AI response reviewed by a human expert before delivery. A separate yearlong Italian study across 31 Salesian schools reports that lessons co-designed with Gemini for Education shifted teaching away from lectures and reduced teachers' reliance on their own energy to hold attention.

How it works

The UK trial was a 10-week randomized controlled trial with over 1,100 students on Eedi's school platform. When a student answered incorrectly, they got one of three kinds of support: a static hint, a human tutor, or interactive AI tutoring. The AI tutor tries to identify the student's specific mistake and guide them to fix it, following a pedagogical prompt supplied by Google. Human experts reviewed each AI response in real time before it reached a student. Teachers also monitored progress and helped in the classroom. Outcomes were tracked on Eedi assignments, intermediate quizzes and independent STAR standardized tests.

The Italian study ran with the Salesians of Don Bosco network and the Pontifical Salesian University in Rome. It covered 1,600 Primary, Secondary and Vocational teachers, representing over 24,500 students aged 13 to 18. It compared classroom activity records and teacher self-reports on lessons planned with and without Gemini.

Evidence

• Expert outright approval of AI responses rose from 76% in the earlier Eedi study to 95%, which Google describes as a pedagogical quality gain of nearly 25%.
• Gemini guided students rather than giving answers in nearly 99% of sessions. Students worked through the math themselves in more than 85% of chats.
• Students averaged about 14 problems per session with AI tutoring against 19 with static hints.
• In Italy, 79% of teachers reported feeling more creative when planning with Gemini, against 30% without. Front-of-class lectures fell 38% and student-led collaborative learning nearly tripled.
• Student-reported engagement rose from 77% to 84% in Gemini co-designed lessons. Teacher reliance on personal energy to keep the room engaged fell 66%. Lessons where activity design drove engagement rose from 23.2% to 52.2%.

Limits

The AI tutor's parity with human tutors was measured with a human expert gating every response. The page gives no result for unreviewed AI output. Gains on in-platform tasks and quizzes did not appear on STAR tests. Google attributes this to slower step-by-step tutoring: students covered less curriculum in 10 weeks than the STAR test spans. That explanation is Google's hypothesis, and full-year follow-up trials are planned but not reported. The Italian results rest largely on teacher and student self-reports and have no randomized design. The blog gives no effect sizes or significance figures for the UK comparison. Those are deferred to a full technical report and to a Don Bosco CNOS-FAP Foundation implementation report, neither of which is reproduced on the page.