Closing the AI guidance gap in education

OpenAI published a report on its education partnerships across Europe, the Middle East and Africa, naming work with governments in Estonia, Slovakia, Greece and Italy. The number that matters comes from ThinkYoung research it cites: 79 percent of surveyed students said they use AI for homework or assignments, but only 47 percent said they had been shown how.

That is a 32-point gap between use and instruction. Students did not wait for permission, and schools are now governing a habit rather than introducing a tool.

We think the core argument here is correct, and it is a little surprising coming from an AI company: as these tools get more capable, teachers matter more, not less. OpenAI says its partnerships rest on equitable access to frontier tools, jointly designed AI literacy programs, and ongoing evaluation adapted to local languages and curricula. Those are the right three pillars.

We also appreciate the hedging. The report calls its findings early, context-specific and not a single model to copy. It says teacher adoption and reported time savings are encouraging, while admitting that whether AI actually strengthens knowledge, independent thinking and retention still needs sustained study. That last question is the whole ballgame, and OpenAI is honest that it has not answered it.

Now the eyes-open part. The report is written by OpenAI's heads of global affairs and education, about programs that put OpenAI's products into national school systems. Teacher time savings are a reasonable early metric, but they measure convenience for adults, not learning for kids. If time savings become the headline success indicator, we risk scaling adoption long before we know whether students are thinking more or less.

The guidance gap also cuts both ways. Closing it could mean teaching students to use AI as a tutor that pushes them to reason. It could also mean normalizing AI as the first stop for every assignment. Which one happens depends on the guidance itself, and on who writes it. We would rather see ministries and teachers set those terms, with vendors supplying tools and data, than the reverse.

Where this leads if it goes right: AI literacy that treats students as critical users rather than consumers. If it goes wrong: a generation fluent in prompting and thin on the underlying skills. The evidence will decide, which is why we want independent evaluation, not just partner reporting.

The report organizes its lessons around five themes: clear educational goals, educator preparation, local partnerships, critical AI literacy and credible evidence. It points to the OpenAI Learning Lab as the vehicle for bringing education researchers together, and references debates around PISA and AI led by the OECD's Andreas Schleicher.