From Chalkboards to Chatbots — World Bank RCT Finds Microsoft Copilot Tutoring Lifts English Scores by Two Years in Edo State, Nigeria

Edo State's Ministry of Education and the World Bank's education team ran a randomized controlled trial testing whether generative AI could accelerate learning in a resource-constrained public-school system, in and around Benin City, Nigeria.

Roughly 800 first-year senior-secondary students across nine public schools were randomly assigned to an after-school English program to test whether teacher-guided use of Microsoft Copilot could accelerate English learning.

The treatment group met twice a week for six weeks and worked through grammar and writing tasks using Microsoft Copilot (running on GPT-4), with teachers introducing each topic and mediating every AI interaction.

Students in the treatment group scored 0.31 standard deviations higher on English assessments than the control group — a gain researchers equated to roughly 1.5 to 2 years of typical schooling in six weeks — and the effect carried over into end-of-year curricular exams covering material beyond the programme itself. Gains were largest among girls and among students who started with higher baseline performance.

The World Bank has been explicit that the 'two years of learning' framing is an extrapolation from a six-week intervention, not a durability claim; independent commentary has stressed the central role of teacher mediation and cautioned against scaling the model before longer-run and unsupervised-use evidence is available.