World Bank RCT finds AI tutoring boosts English learning outcomes in Nigeria
800 Nigerian students using GPT-4-based Copilot for six weeks scored 0.23 SD higher on English, equivalent to about 1.5 to two years of ordinary schooling.
- Culture & impact
- Notable
The World Bank published results from a randomised controlled trial testing whether generative AI could improve English-language learning in Nigerian secondary schools, reporting a statistically significant effect after a six-week pilot.
Between June and July 2024, 800 first-year senior secondary students attended twice-weekly after-school English classes in school computer labs. A teacher introduced each week’s topic before students worked through grammar and writing exercises with Microsoft Copilot, built on GPT-4. Students in the programme scored 0.31 standard deviations higher than a control group on a combined assessment spanning curriculum-aligned English, AI literacy and digital skills, and 0.23 standard deviations higher on English specifically, the trial’s primary outcome. The World Bank’s cost-effectiveness analysis put the gain at roughly 1.5 to two years’ worth of “business-as-usual” schooling for the cost incurred, placing it among the more cost-effective learning interventions the Bank has evaluated. Effects were largest for female students and for those with stronger baseline performance, rather than evenly distributed across the ability range.
The finding was announced in a research brief and video in February 2025, with the full Policy Research Working Paper following in May. It stood out because much contemporaneous research on generative-AI tutoring in wealthier countries had reported null or mixed effects on standardised outcomes; a rigorously randomised trial with a clearly positive effect size, run in a lower-income setting using a widely available commercial product rather than a bespoke research tool, was comparatively rare in the literature at the time.
The World Bank framed the result as encouraging for scaling AI-assisted instruction in contexts with severe teacher shortages, while noting the pilot was short, supervised rather than fully autonomous, and dependent on computer-lab infrastructure and connectivity that many schools in the region lack.