AI Proves It Can Give Students 1.5 Extra Years of Learning — And More of This Week's Biggest EdTech Stories

Google's Gemini AI Gives Sierra Leone Students 1.5 Extra Years of Learning in 8 Weeks

Google has released results from a pre-registered randomised controlled trial that is, frankly, staggering. Working with Fab AI and local teachers across 48 mathematics classrooms in Sierra Leone, researchers randomly assigned nearly 1,800 Grade 7 and 8 students to either use Gemini's "Guided Learning" mode or continue with standard teaching. After just eight weeks, the students using Gemini improved their test scores by +0.26 standard deviations — the equivalent of 1.2 to 1.7 years of typical learning progress in a low-to-middle-income country context.

Students who hit the recommended 12-hour usage threshold did even better, recording a +0.38 standard deviation gain — enough to move an average student from the middle of their class all the way into the top third. Crucially, engagement was high: the average student clocked 15 hours of AI use, well above the target. This is not a lab study with cherry-picked participants — it is a real-world RCT with all the rigour that implies.

Alongside the Sierra Leone data, Google also published results from a separate Italian study across the Don Bosco school network — 700 educators and 9,000 students spanning primary to vocational college. Teachers using Gemini for lesson planning and content creation reported that 80–99% of students in each class successfully mastered their planned lesson skills. Teachers also reported a 70% reduction in administrative time, which they reinvested directly into one-to-one student mentorship. Google simultaneously announced expansions of its AI Educator Series into India and a new partnership with the African Union Commission covering all 55 member states.

The Sierra Leone data alone could reshape how development organisations think about education investment. This is the moment AI in education stopped being a promise and became a proof point.