Not Faster, but Better — Peru's Randomised Trial of One-Day Copilot Training for 1,269 Lima Teachers

Between March 2025 and 2026 the World Bank's Development Research Group, Lima's regional education directorate (DRELM) and Microsoft tested whether a short, practice-oriented course could help teachers adopt generative AI. 1,269 sixth-grade teachers in 390 public schools in Metropolitan Lima took part: 196 schools (660 teachers) were assigned to treatment and 194 schools (609 teachers) to control.
The intervention combined one day of in-person training (co-designed with teachers and delivered by university lecturers), access to Microsoft Copilot, and eight months of follow-up through WhatsApp. Teachers were taught three rules: treat AI as an assistant, verify its output, and protect student privacy.
Results over the 28-week programme: Copilot activation rose by 20.4 percentage points and treated teachers accumulated 3.15 more active weeks on the platform. Prompt-writing quality improved by about 9.2% (programme assessment) and 11.1% (independent assessment); teachers were 9.9 pp less likely to misjudge weak prompts and 6.7 pp less likely to accept inadequate AI responses.
Cautionary findings: the study found no significant time savings across seven professional activities (including lesson planning, grading and parent communication), and it measured teacher capability, not student achievement, so whether the skill gains improve learning is still unknown. Many teachers also work from home because about 65% of Peruvian primary schools lack internet access. The trial supports pairing AI access with skills training rather than distributing licences alone.

Read the full analysis: https://documents.worldbank.org/en/publication/documents-reports/documentdetail/099309210012625906