AI Tutors in African Classrooms: Experiment or Transformation?
In a classroom in Benin City, a teenager named Omorogbe Uyiosa sat with his phone after school hours, working through English exercises with a generative AI system that answered his questions and adjusted to how he wrote. Uyi’s school, Edo Boys High School, was one of dozens in Edo State that took part in a World Bank-backed pilot testing whether AI could function as a virtual tutor in a public school system with limited staff and limited time. The results, published by the World Bank, showed that students who went through the six-week after-school program outperformed their peers not only in English, the pilot’s main focus, but on end-of-year exams covering material the program never touched.
That single data point has become one of the most cited pieces of evidence in a debate now playing out across the continent: is AI tutoring a genuine answer to Africa’s learning crisis, or a well-funded experiment that looks better in pilot form than it will at scale.
Why the Question Matters Now
Africa’s education systems have long struggled with a mismatch between enrollment and actual learning. Millions of children attend school but leave without basic literacy or numeracy, a problem education economists call learning poverty. Class sizes in many public schools run into the dozens, sometimes past a hundred pupils per teacher, leaving little room for individual attention. It is this gap, between children who are present in school and children who are actually learning, that AI tutoring tools are being pitched to close.
The pitch has attracted serious institutional weight. In November 2025, more than a hundred education officials, funders, and technology executives gathered in Nairobi for the AI for Education Summit, organized around what the Gates Foundation’s Ben Piper called “grounded ambition” — a deliberate pushback against hype, paired with an acknowledgment that the technology genuinely has something to offer under-resourced systems.
What Is Actually Being Deployed
The Edo pilot is not an isolated case. In Kenya, a structured pedagogy platform called EIDU is now used by close to 400,000 children, with the World Bank citing demonstrated learning gains from the program. Ghana has tested AI tutoring delivered through WhatsApp, a deliberate choice given how widely the app is already used on ordinary phones. Similar adaptive learning pilots are running in Côte d’Ivoire, the Gambia, and Mali, while Ethiopia has focused its efforts on tools that support teachers rather than students directly.
In South Africa, a WhatsApp-based tutor called Luma Learn has reached more than ten thousand learners, offering explanations in isiZulu and other local languages. Its founder, Chris Folayan, built it on WhatsApp specifically because it requires no new app and works on low bandwidth, addressing two of the most common barriers to digital learning tools on the continent: device cost and data cost.
Rwanda has taken the most explicit government-to-government approach. In late 2025, the Rwandan government partnered with Anthropic and the pan-African training organization ALX to bring AI tools into as many as 2,000 classrooms, training teachers directly rather than routing everything through students. The partnership is framed as part of Rwanda’s Vision 2050 push to build an AI-literate workforce, and it reflects a broader pattern: several of the more credible deployments treat AI as a teacher’s assistant first, and a student-facing tutor second.
The Policy Backdrop
None of this is happening in a vacuum. The African Union adopted its Continental Artificial Intelligence Strategy in July 2024, explicitly naming education as one of the sectors expected to benefit from AI adoption, while also flagging the infrastructure gaps, electricity, connectivity, and skills, that stand in the way. Individual governments, Nigeria and Rwanda among them, have since moved to draft their own national AI strategies and, in Nigeria’s case, education-specific frameworks addressing how schools and examination bodies should treat AI-assisted work.
That last point matters because the same tools narrowing learning gaps are also complicating assessment. Exam bodies and universities across the continent are grappling with AI-assisted cheating even as their own ministries fund AI tutoring pilots, an inconsistency that policymakers have not fully resolved.
What the Evidence Does and Does Not Show
The strongest results so far, the Edo State randomized evaluation in particular, come from short, tightly supervised pilots with teacher involvement built in. That is a meaningfully different proposition from unsupervised access to a chatbot, and it is a distinction the researchers behind these programs are careful to draw. What remains unanswered is what happens after the pilot funding ends, whether gains persist over a full academic year, and whether results replicate in schools without the accompanying teacher training and monitoring that made the initial pilots work.
A Measured Middle Ground
The honest answer, for now, is that AI tutoring in African classrooms is neither a settled transformation nor a passing experiment. It is a set of promising, narrowly evidenced interventions that work best when paired with teacher support, local language design, and infrastructure realistic for the setting, features present in Edo, Kigali, and the WhatsApp-based programs, but not guaranteed in every deployment chasing the same headline results. Whether that combination scales beyond pilot budgets, across thousands of schools rather than hundreds, is the question the next few years of implementation will actually have to answer.
