By Jawad Asghar and Clio Dintilhac, Gates Foundation

In a Grade 3 class in Gujarat, India, a teacher runs a 60-second oral-reading check using her basic mobile phone. Errors are flagged instantly, and the AI suggests how to regroup the learners based on an instantaneous analysis of the reading check. By the next lesson period, the teacher has reorganized the students for a short reading fluency lesson and the app reminds the teacher of a few simple follow-ups she already knows how to run. Because the tool works offline and syncs later, the routine is reliable, even without steady internet connectivity.

This real-life example shows the potential of Generative AI (GenAI) to expand what proven approaches, such as structured pedagogy and targeted/remedial instruction (e.g., Teaching at the Right Level), can already do well, and to ease the pain points these approaches have faced in scaling up. In practical terms, it can enable the tried and true diagnose – group – practice loop to run smarter, simpler, and faster. And it is already in practice: in just over a year, Wadhwani AI’s version of this tool has logged more than 3.3 million assessments, showing that these workflows are already working in thousands of classrooms in India and that they provide powerful, practical support for teachers, not substitutes or workarounds for them.

Starting with teachers

When GenAI is used well in classrooms, it accelerates what effective teachers already do in structured pedagogy and targeted/remedial instruction. It speeds lesson prep, such as with lesson scripts, vocabulary-leveled texts, and end-of-lesson checks. It turns these quick checks into same-day regrouping suggestions. It offers just-in-time prompts for teachers in the language learners speak at home, aligned to curriculum, while teachers stay in charge of decisions. The result is less time spent by teachers on administration and more time on instruction. Our background paper with the support of Woodspring Advisory and the inputs of Sara Cohen, Amber Gove, and Terri Horn, informed the 2025 Spotlight report on foundational learning in Africa.describes four use cases across the globe. It underlines the strong potential, as GenAI evolves, to ease the practical bottlenecks that have limited scaling while keeping teacher judgment at the centre.

What this looks like in practice

On the teacher-facing side, current tools already generate aligned lesson outlines and materials, propose groups from quick checks, and provide brief, practical coaching tied to what the teacher is seeing. Examples include classroom grouping and classroom support tools, and planning tools that help tailor activities to different levels. These act as a teaching partner that keeps pedagogy front and centre.

Alongside teacher-facing tools, the background paper sets out three other use cases: student-facing supports that deliver offline-first practice and voice-based reading checks in local languages, system-facing tools that create secure data flows and simple dashboards that work with education information systems, and ecosystem support for shared standards, benchmarks, procurement guidance, and privacy.

Teacher-in-the-loop models are emerging across sub-Saharan Africa. In Nigeria’s Edo State, teachers led twice-weekly sessions where learners practiced with an AI assistant using structured prompts. Teachers selected and adjusted prompts, reviewed student work, and kept the class on task. A major study found meaningful learning gains in only six weeks, with girls benefiting the most. This suggests that GenAI and teachers working in tandem can deliver fast, practical progress in public schools.

For classrooms without one-to-one devices or steady connectivity, low-bandwidth services show another route. In Ghana and Sierra Leone, Rising Academies schools run Rori, a free tutor that provides brief mathematics practice over WhatsApp during and after school, aligned to early-grade skill progressions. Teachers can quickly see a concise summary for each learner that shows what they worked on and where they struggled, then decide which skills to revisit or how to group learners. This echoes the same diagnostic and regrouping routines used in remedial or targeted instruction, such as Teaching at the Right Level.

Eight principles for introducing GenAI safely and smartly

Applying GenAI for education in sub-Saharan Africa requires more than tools. It requires a principled approach. The eight principles below offer a practical way to move from scattered pilots to sustainable, system-level gains. They are anchored in regional realities: uneven (or no) connectivity, multilingual classrooms, limited teacher time, strict child-data privacy requirements, fragmented government data systems, and persistent equity gaps. Applied together, these principles aim to keep GenAI useful today, ready for tomorrow, and aligned with countries’ national goals for progress.

1. Start offline. Core functions, including assessment, regrouping, and practice, should run on basic devices. Use connectivity to sync results and update content.
2. Build for African languages. Support the languages children and teachers actually use, including the way children and adults often mix languages in the same sentence. Keep accuracy and relevance under the care of local teams.
3. Keep teachers in the lead. Make AI suggestions optional and easy to adjust. Let teachers choose when and how to use them, approve changes to grouping or pacing, and override the tool at any time. Provide simple training and coaching so teachers are confident, and make it clear that professional judgment takes precedence.
4. Design for interoperability. Use open standards and shared data models so education information systems, assessments, and classroom tools work together in national systems and so countries can reduce the risk of lock-in.
5. Build on African expertise. Invest in and rely on the depth of African technical, linguistic, and pedagogical expertise so adaptation is continuous, relevant, and locally led.
6. Strengthen evidence, quickly and well. Broaden evaluation approaches to get decision-grade evidence faster. Consider evaluation as a spectrum, from well-designed randomized trials to operational classroom pilots. The middle ground, including quasi-experimental and mixed-method studies, helps explain not only if something works but why. Together, these approaches build confidence early and strengthen evidence as tools mature.
7. Put equity first. Plan for shared devices, print options, and low-bandwidth choices such as text messaging or widely used messaging apps such as WhatsApp when connections are weak, so no learner is left out.
8. Protect children’s privacy and data. Use data minimization, clear consent for families and teachers, and strong security. Give learners and families the ability to see what is collected, correct it, and take the data with them. Provide options to use tools that deal with limited or no personal data.

Moving from promise to practice

The choice is not between ‘doing more of the same’ in classrooms and a leap into the unknown. We already know what moves the needle in reading and numeracy. These principles point to a practical pathway for using GenAI to ease the everyday bottlenecks that hold back effective teaching and system management so that proven approaches to literacy and numeracy can reach more classrooms and last. The task now is to back African educators, researchers, and governments to lead this work, so that GenAI in education grows on their terms and delivers tangible gains for every learner.