Measuring the impact of AI on teaching and learning

Source: Google Date: 2026-05-19 URL: https://blog.google/products-and-platforms/products/education/measuring-the-impact-of-ai-on-teaching-and-learning/

Summary

Google published two RCTs measuring Gemini in classrooms: a Sierra Leone study (1,800 students, 8 weeks) showed +0.26 SD learning gains — roughly 1.2–1.7 years of typical progress — rising to +0.38 SD at 12+ hours of use. An Italy study (700 educators, 9,000 students) reported 80–99% lesson mastery rates and a 70% reduction in teacher administrative time. Alongside the data, Google announced expansion programs: an AI Educator Series launching in six Indian states with localized language support, and an African Union partnership bringing Gemini for Education and NotebookLM to 55 nations.

Implications

This signal sits at the edge of the AI ecosystem thread rather than core agentic engineering, but its relevance is structural: Google is publishing RCT data — a higher evidentiary bar than most vendor AI claims — and using it to justify geographic rollout at scale. The effect sizes are large enough to be credible signal rather than marketing noise.

• AI ecosystem/power dynamics: Google is positioning Gemini as measurably superior for educational deployment, giving it a differentiation story against generic LLM access. If the RCT methodology holds up to scrutiny, this becomes a replicable playbook for institutional AI adoption.
• Watch: whether independent researchers replicate these studies, and whether the “hours of use” dosage effect (~12h threshold) appears in other domains — it would suggest a general “minimum viable exposure” curve for AI-assisted skill acquisition.