From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria

Summary

A teacher-supported six-week programme improved paper-test outcomes, with important attrition and design caveats.

Volunteers in nine schools were randomized to a teacher-supported Copilot programme or ordinary schooling. The report estimates 0.23 SD in English and 0.31 SD on a composite, measured by supervised paper tests.

Of 1,328 randomized students, 759 completed the final assessment; authors examine attrition sensitivity. The intervention bundles AI, instruction, practice and infrastructure, preventing attribution to chatbot access alone.

Working paper; short-term outcomes, volunteer participation and substantial attrition. Schooling-equivalent years are a conversion, not observed retention.