OECD warns AI boosts grades but not lasting learning

Zülal Atagün – ISTANBUL

The Organization for Economic Cooperation and Development (OECD) has warned that while generative artificial intelligence can boost students’ grade in the short term, it does not necessarily translate into lasting learning, according to its newly released Digital Education Outlook 2026 report.

Drawing on recent surveys across multiple countries, the report finds that general-purpose generative AI tools help students complete tasks faster and more accurately, leading to measurable short-term performance gains.

However, the OECD cautions that outsourcing cognitive effort to AI during learning may limit long-term skill development and weaken knowledge retention.

A field study conducted in Türkiye involving around 1,000 high school students illustrates the pattern. Students with access to general-purpose generative AI tools saw their short-term performance in mathematics exercises rise by 48 percent.

Those using a pedagogically designed “GPT tutor” version recorded an even sharper short-term increase of 127 percent.

Yet when AI access was removed and students were tested again, their scores fell below expectations. Students who had previously used AI performed up to 17 percent worse than peers who had never had AI access. The OECD concludes that higher short-term performance does not automatically indicate durable learning.

Similar studies in Canada, France, Sweden and the Netherlands produced comparable results.

The OECD notes that AI tools designed specifically for learning purposes can support more durable learning by providing guided questioning, hints and strategy prompts.

Türkiye is identified among countries where students make intensive use of AI outside school, mainly for explanation of topics, homework solutions, personalized learning plans and progress tracking.

A 2025 European study cited in the report shows that 10 out of 23 countries, including Türkiye, have formally incorporated generative AI into their education strategies.

Commenting on the findings, Yelkin Diker Coşkun of Yeditepe University’s Education Department emphasized that general chatbots are not designed to teach and warned that overreliance may turn students into passive users rather than active learners, particularly in tasks requiring higher-order thinking, synthesis and the ability to transfer knowledge to new problems.