OECD on AI in Education: Performance Is Not Learning

In short: The OECD's Digital Education Outlook 2026, published on February 3, 2026, draws one line that matters for every school and parent: students who lean on general-purpose generative AI can improve their task performance without matching learning gains, and in one trial did worse once the AI was taken away. The OECD says generative AI can support learning when it is designed to, and can weaken skills when used as a shortcut. The report also points to real time savings for teachers, and it warns that access is unequal.

Key Takeaways

• The OECD says overreliance on tools that give direct answers can improve task performance "without corresponding learning gains".
• It links this to "metacognitive laziness", where handing thinking to a tool reduces long-term skill building.
• Tools designed to support learning, such as intelligent tutoring systems, fare better than answer-giving chatbots, according to the OECD.
• In the OECD's 2024 TALIS survey, 37% of teachers said they use AI in their work, and about seven in ten believe AI could let students present others' work as their own.
• The report warns that a digital divide could widen, and discusses tools that work offline or with limited connectivity.

What Is the Evidence?

The source is the OECD Digital Education Outlook 2026, a report that reviews emerging research on generative AI across three settings: students learning on their own, students and teachers together, and teachers working alone. All figures below come from the report itself.

The report's key points are:

• Students who rely on tools that give direct answers can improve task performance "without corresponding learning gains", and generative AI "can displace cognitive effort and weaken the skills that underpin deep learning" when used as a shortcut.
• The OECD calls this mechanism "metacognitive laziness".
• In a field experiment in Türkiye, access to GPT-4 improved short-term performance by 48% with the standard interface and by 127% with a tutoring version designed to support learning, but students performed 17% worse once access was removed.
• Teacher use is already common: according to the OECD's 2024 Teaching and Learning International Survey (TALIS), 37% of teachers use generative AI for work tasks, with large variation across countries. About 64% of teacher AI users use it to generate lesson plans, and around 50% agree it helps them create or improve lesson plans (from 18% in France to 91% in Viet Nam).
• On integrity, seven in ten teachers on average believe AI could let students misrepresent others' work as their own.

For teachers, the report cites a 31% reduction in time spent on lesson and resource planning by secondary science teachers in England, and a 9 percentage-point increase in student pass rates when low-experience tutors used AI support, with smaller gains for more experienced tutors. The Türkiye field study is the high school trial we cover in AI tutoring and personalized learning: what the trials show.

What Does the Technology Do in a Classroom?

Generative AI produces text, explanations and feedback on demand. The OECD says it can scale personalised learning through intelligent tutoring systems, including in low-infrastructure settings. In practice there are two very different uses. A student can ask a chatbot for the answer, which finishes the task. Or a tool can be built to ask questions, give hints and make the student do the reasoning. The OECD's finding is that the first use raises performance and the second is more likely to raise learning.

What Changed?

The change is in how the question is framed. Early AI-in-education talk asked whether students could produce good work with AI. The OECD's framing is that producing good work and learning are different outcomes, and the second one is measured when the tool is gone. That is why exams, or tests taken without AI, are the revealing measure in the studies it reviews.

Who Benefits?

According to the report, teachers see the clearest near-term gain, in the form of time saved on planning. The reported support for less-experienced tutors is another bright spot. Students benefit when the tool is designed to teach, with intelligent tutoring systems highlighted as promising in the report. For a rural or low-connectivity context such as Nepal, the benefit depends on access. The OECD calls for "inclusive GenAI infrastructure for all", which it says needs devices, connectivity, curriculum-aligned resources and professional development. Our look at open AI models for Nepali language and low connectivity covers one part of that problem.

What Are the Limitations?

• Most evidence is short-term. The studies reviewed measure outcomes soon after the intervention, so long-run effects are not established.
• Teacher-time figures are for specific groups. The 31% planning saving is for secondary science teachers in England, and may not carry over elsewhere.
• Integrity is a live concern. 72% of teachers in the TALIS survey worried that students could present AI work as their own.
• Self-selection and context. Teachers who adopt AI early may differ from those who do not.

What Happens Next?

This is expectation, not a finding. The OECD's direction suggests schools will be pushed toward tools designed for teaching, rules about when AI is allowed, and assessments that show what a student can do without it. Watch for more controlled trials in lower-income settings, where evidence is sparse. See also how AI is changing everyday life: what usage data shows.

The Short Version

The OECD's message is that a student who finishes an assignment with AI has not necessarily learned anything. AI designed to teach, with teachers in charge and fair access, can help. AI used as an answer machine can leave learners worse off once it is removed.

Related Insights

• AI Tutoring and Personalized Learning: What the Trials Show
• How AI Is Changing Everyday Life: What Usage Data Shows
• Open AI Models for Nepal: Nepali Language and Low Connectivity
• Why AI Matters for Nepal: What the Evidence Shows

Sources

• OECD, OECD Digital Education Outlook 2026: Exploring effective uses of generative AI in education, February 3, 2026