OpenAI's education report documents a guidance gap, not a learning gain

On 10 November 2026, OpenAI published a signed post, "Closing the AI guidance gap in education," by Tom Duff Gordon, vice president for Global Affairs in EMEA, and Leah Belsky, vice president for Education. The checkable finding is a guidance gap, not a learning gain. OpenAI cites ThinkYoung research that 79% of surveyed students reported using AI for homework or assignments, while only 47% reported being shown how. The company's account of its education partnerships in Europe, the Middle East and Africa still treats teacher adoption and reported time savings as early success signals. It also says, plainly, that whether AI strengthens knowledge, independent thinking, and retention requires sustained evaluation, and that the report is not a single model other school systems can copy.

What the numbers actually say

Two sets of figures need to be kept apart. The 79% and 47% contrast is what OpenAI cites from ThinkYoung, and it is the only student-use comparison printed on the OpenAI page we could read. It describes a gap between students already using AI for schoolwork and students who say they were shown how to do so. It does not measure whether those students learned more, remembered more, or thought more independently.

The sample description comes from ThinkYoung's own page, "Views, Experiences & Expectations for AI in Education." That page says the study draws on surveys with 5,000 high-school students aged 13-19, together with 600 teachers and a series of focus groups with students and parents, across 12 countries in Europe, the Middle East and Africa. It also prints a large figure that 95% of teens use AI tools. That 95% figure is a broader claim about teen use of AI tools. It is not the same as the 79% figure OpenAI cites for homework or assignments. OpenAI's post does not itself print the 5,000, 600, 12-country, or 95% numbers.

ThinkYoung's page asks visitors to complete a form to access the full research report. We were not able to read that PDF, so we cannot report survey dates, question wording, response rates, or how the sample was drawn. Those details may be in the gated report. They are not in the public pages we could verify.

What OpenAI reports from its partnerships

The post is also a report on OpenAI's education partnerships in EMEA. It says the company is working with governments and education systems, including in Estonia, Slovakia, Greece and Italy, among others. Those collaborations, it writes, share three priorities: equitable access to frontier AI tools, jointly designed AI literacy programmes, and ongoing measurement and evaluation, while adapting to local languages, curricula and institutions.

Five themes organise the lessons: clear educational goals, educator preparation, local partnerships, critical AI literacy, and credible evidence. Experiences in Estonia, Greece and Slovakia are said to point to public leadership that sets a clear direction, investment in teachers' skills and confidence, and practical support tailored to local curricula and classroom needs. OpenAI says implementation will vary across countries, that country examples reflect particular contexts and arrangements, and that findings should be interpreted within those limits. "These are early experiences, rather than a single model for education systems to follow," the authors write.

The post also says OpenAI is learning how to judge progress with evidence. Teacher adoption and reported time savings "offer encouraging early signals." Surveys, programme monitoring and consultation are described as ways to understand participants' experiences and the barriers they encounter. The text we could read does not give numbers for adoption or for time saved, and it does not name independent evaluations of student learning.

OpenAI refers to recent debates around PISA, the OECD's international student assessment, and AI, including those led by the OECD's Andreas Schleicher, and then states a lesson: as AI becomes more capable, the role of teachers becomes more important, not less. We were not able to fetch and read Schleicher's remarks independently, so we treat that lesson as OpenAI's framing of those debates, not as a verified quotation from Schleicher.

The fair case

The fair reading is that a guidance gap can be a real problem even if learning gains have not been shown. If a large share of surveyed students already use AI for homework or assignments, and only a minority say they were shown how, then teacher preparation and clear rules of use are a bottleneck, not a side issue. Schools cannot wait for perfect learning studies before they decide how the tools already in students' hands should be taught.

Time saved by teachers can also be a legitimate early operational metric. If a partnership is still new, it is reasonable to ask first whether teachers will use the tools and whether the tools free up hours, while studies of knowledge, independent thinking, and retention are still being designed. OpenAI is explicit that those harder questions remain open, and that it has a responsibility to contribute evidence, including when the findings challenge its assumptions. It points to the OpenAI Learning Lab as a place where education researchers are being brought together to study when and how AI can support education systems.

What is missing

What the public post does not provide is as important as what it does. It does not publish teacher-adoption rates or time-saving figures, even as it calls those signals encouraging. It does not give survey dates or the ThinkYoung instrument. The "below we set out" line on the page suggests further practical guidance, but we did not obtain a longer report body or a separate PDF with country-level results, so we do not invent those details.

The 79% and 47% figures are reports of what students said, from a study whose full methods we could not inspect. They should not be read as a global rate, or as evidence that guidance, once provided, improves learning. The 95% teen-use figure on ThinkYoung's page is likewise a page-level claim we cannot audit further without the gated report. Partnership lessons from Estonia, Slovakia, Greece and Italy are, by OpenAI's own account, context-specific.

Who should care, and what to watch

Education ministries, school leaders, and teachers already dealing with unofficial student use of AI should care, as should anyone inclined to treat a vendor report as proof that classroom AI works. The useful immediate fact is the gap between use and guidance, as cited. The useful limit is that learning outcomes are still an open evaluation question.

In the next six months, the checks that matter are specific. Does ThinkYoung release the full report in a form that can be read without a gate, including dates, questions, and sampling? Does OpenAI publish the teacher-adoption and time-saving figures it already treats as early signals? And does the OpenAI Learning Lab, or an independent study, produce evidence on knowledge, independent thinking, and retention, including results that cut against the company's hopes?

Close

OpenAI has documented a guidance gap and has been careful, in the same text, not to present its EMEA partnerships as a finished model or as proof of learning gains. That honesty should be taken at face value. The next public test is whether the missing methods and the missing learning measures appear, or whether the 79% and 47% contrast is left to stand in for evidence it cannot provide.