From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria
Upper secondary AI and learning Reading and writing
Links to claims
Limits (1)
| Claim | Basis |
|---|---|
| C2 | Nigeria, senior secondary, student-level RCT (working paper): teacher-guided after-school Copilot (GPT-4) sessions for six weeks raised the programme test by 0.31 SD and the school's own term English exam by 0.21 SD. Guided use with extra time, not homework outsourcing; confounded with extra instruction time. |
Review state: first-pass draft, pending human review.
Source note
Syntes: De Simone et al. (2025) — teacher-guided Copilot sessions after school raised English results in Nigeria (working paper)
Källa: De Simone, M.; Tiberti, F.; Barron Rodriguez, M.; Manolio, F.; Mosuro, W.; Dikoru, E. J. (2025 (World Bank Policy Research Working Paper 11125, May 2025)). From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria. http://hdl.handle.net/10986/43212
Verified numbers (full text read 2026-10-11, World Bank PDF)
- Setting: Benin City, Nigeria; nine schools with computer labs; first-year senior secondary students (typically 15). Six weeks, June–July 2024, up to two 1.5-hour after-school sessions per week.
- Intervention: students in pairs used Microsoft Copilot (GPT-4 at the time) for English topics in the national curriculum; each session started with a teacher-provided prompt, then free interaction; teachers trained three days and circulated to keep students on task.
- Design: among interested students, simple random assignment: 657 treatment, 671 control (control had no programme). Final assessment completed by 422 treatment and 337 control (attrition differs; authors report bounds and say conclusions hold).
- Results (ITT, controlling for second-term exam and school): total score 0.31 SD (SE 0.068), IRT-scaled 0.263 SD; English 0.238 SD, AI knowledge 0.309 SD, digital skills 0.139 SD (10 % level). The school's third-term English exam (whole term, not only the programme content): 0.206 SD (SE 0.067).
- Caveats (authors and design): the treatment group got extra instruction time (after school) that the control group did not, so AI is not separated from extra time and teacher attention (the authors suggest a future arm with teacher tutoring without technology); some control students attended sessions; working paper.
Why it is in the library
A school-age randomised trial with an outcome measured without the tool (the school's term exam) in a low-resource system. It bears on C2: teacher-guided use of an answer-capable chatbot did not lower, and here raised, results on an unaided exam. It is not a test of unguided homework outsourcing, and the extra-time confound means it is not a clean AI-versus-no-AI contrast.
Metadata as recorded
- ID
- AI-024
- Tier
- RCT
- Closed-book follow-up
- partly (programme-aligned paper-and-pencil multiple-choice test at the end; also the school's own third-term English exam, without the tool)
- Titel
- From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria
- Författare
- De Simone, M.; Tiberti, F.; Barron Rodriguez, M.; Manolio, F.; Mosuro, W.; Dikoru, E. J.
- År
- 2025 (World Bank Policy Research Working Paper 11125, May 2025)
- Källa/DOI
- http://hdl.handle.net/10986/43212
- URL
- https://documents.worldbank.org/en/publication/documents-reports/documentdetail/099548105192529324 ; RePEc: https://ideas.repec.org/p/wbk/wbrwps/11125.html
- Typ
- student-level randomised trial of a six-week after-school programme (two 1.5-hour sessions a week, June–July 2024) in which first-year senior secondary students (about 15 years old) in nine schools in Benin City used Microsoft Copilot (GPT-4) in pairs for English, with teacher-provided prompts and teachers circulating; control: no programme, regular classes
- Åldersgrupp
- gymnasium
- Teman
- ai-och-larande;lasning-och-skrivande
- Hämtad
- 2026-10-11
- Paywall
- open (CC BY 3.0 IGO)
- Tillagd
- source sweep 2026-10-11 (kind: new); links pending human review
- Preprint
- World Bank Policy Research Working Paper, not peer reviewed as of 2026-10-11 (applies if used in a verdict).