OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education

Motivation and Scope

These Insights on the effective use of generative AI in education aim to contribute to discussions about the use of generative AI (GenAI) in education, based on emerging evidence as well as expert knowledge on the use of artificial intelligence (AI) in education. They are intended to empower jurisdictions, teachers, students, parents and other education stakeholders with easily accessible knowledge to responsibly seize the opportunities of GenAI. GenAI tools are defined as AI systems that create new outputs (e.g. text, code, audio, images, video), often in response to prompts, based on their training data. The ECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education presents the available evidence about challenges and promises of GenAI in education. This knowledge base has been supplemented by a broad country and expert consultation.

While most people in education see AI and generative AI as identical, GenAI is a subset of AI, and, in the case of education, using generative AI tools sometimes has different consequences from using non-generative AI tools. While still limited, emerging research and evidence on the effects of the use of GenAI on teaching and learning allows for a more specific focus than the previous OECD-Education International (2023) guidelines on AI in education.

While references are not explicitly mentioned for the sake of legibility, most of the evidence underlying this document comes from the OECD Digital Education Outlook 2026 (and its previous 2023 and 2021 editions as far as policy and broader AI considerations are concerned). Some additional OECD references are provided as an Annex. They point to the broader OECD knowledge base on AI that informs the document.

These Insights cover both the school and higher education sectors, including vocational education and training, and go beyond issues that are mainly relevant to teachers and institution administrators. To make the language lighter, “teachers” refers to instructors at all levels of education, including higher education (and “academic staff” is used when only higher education faculty members are concerned) and “students” refers to both school pupils and higher education students.

Section 2 provides an overview of the Insights and presents the opportunities and challenges of GenAI in education. Section 3 provides more detailed guidance and explanations to make the adoption of GenAI in education a success.

Opportunities and challenges of generative AI in education

To most people AI corresponds to a general-purpose technology mainly experienced via chatbots based on large language models (LLMs) such as OpenAI’s ChatGPT, Google’s Gemini, Microsoft’s Co-pilot, Anthropic’s Claude or Mistral’s LeChat. But AI tools are manifold. Whereas non-generative AI tools mainly produce predictions, classifications, recommendations and ratings, for example of movies, books or other products and services, generative AI (GenAI) tools produce new outputs such as text, pictures, videos, songs, mathematical equations, computer programmes, often following a question or command (“prompt”), based on their training data. Text generation allows LLMs to be very versatile chatbots and have fluid, real-time dialogue-based interactions.

Beyond being generative or predictive (or rule-based), AI tools can also be general-purpose or educational (or specialised for research or other goals) (Table 14.1):

• Non-generative general-purpose AI tools include text-to-speech and speech-to-text systems or note-takers: they are used in education but could be used for any other purpose; • Non-generative educational AI tools include intelligent tutoring systems or early warning systems; • Generative general-purpose AI tools include LLMs such as ChatGPT; • Generative educational AI tools include LLMs developed for educational purposes, such as Khanmigo or the Socratic Playground, that are finetuned or configured to refrain from giving solutions, and support students to get themselves to the answer or solution to their problem.

These distinctions are important to understanding the opportunities and challenges of generative AI in education. However, the appropriate use of either educational or general-purpose GenAI matters even more. While educational tools are merely defined as having the objective of supporting student learning or teachers’ teaching, they are usually expected to also have other characteristics, including reinforced safety requirements, curricular alignment, age appropriateness and pedagogical control mechanisms.

Adoption of GenAI in education

While non-generative educational AI tools have remained largely invisible to the general public (and not widely utilised in education), especially before the COVID pandemic, the emergence of GenAI has made AI and its possibilities visible to all and challenged teachers and the education sector. Contrary to most educational AI, general-purpose GenAI tools are available for free outside of school and students and teachers can (and do) use them whether they are provided by their universities or schools or not. Contrary to most other non-generative educational AI tools, most GenAI tools are not designed to support students’ learning, but to perform tasks, including learning tasks that students typically receive from teachers as educational assignments. As a result, the first response to the emergence of LLMs was an outcry about students’ academic integrity and cheating. While a small market for the outsourcing of educational assignments to humans existed before GenAI, this possibility suddenly became freely accessible to most. This first reaction mainly spoke to the “credentialing” and “sorting” function of education systems, that have to guarantee the reliability, fairness and comparability of grades, degrees and competence.

Chatbots (e.g. ChatGPT, Deepseek, Gemini) Image, video or sound generators

As long as they have an adequate device and connectivity, students and teachers can use powerful general-purpose GenAI tools. Domestic and international surveys show that students already very widely do so outside of the classroom, and more so as they get older. They may access GenAI in very different ways in different contexts though, and even run GenAI tools that work with no connectivity. A sizeable share of their parents also uses GenAI, for work and personal reasons. And teachers use GenAI tools to generate lesson plans, quizzes, assignments, and even sometimes to grade and provide feedback to their students. In some cases, they may use specialised educational GenAI tools. Academic staffand other scientists also increasingly use GenAI for research, often just to improve the fluidity of their English as a foreign language, but sometimes also to solve problems that would be impossible to solve quickly (or solve at all) without GenAI tools. Specialised GenAI tools have been developed for most stages of the research process, especially for natural sciences. GenAI has now become part of the social and educational reality and is therefore difficult to ignore. Given the pervasiveness of GenAI in the labour market, many call for education to foster GenAI literacy as a way to upskill and reskill students and professionals to effectively participate in the workforce.

Strengths and weaknesses of GenAI systems

The scientific achievement of a specialised scientific generative AI system, AlphaFold, gives a good example of the possibilities of GenAI tools: within a year, the model generated the 3D shape of 200 million proteins based on their molecular composition, virtually all that were known so far. Its inventors were awarded the Nobel Prize in chemistry in 2024. Before then, it took (usually a PhD student) three years to generate the shape of a single protein. This scientific breakthrough has opened new possibilities in medicine to develop drugs and refocused human research on more complex scientific problems in structural chemistry. GenAI tools go beyond helping us to retrieve knowledge, they also contribute to knowledge production and productivity.

What is so special about general-purpose GenAI tools, which are typically experienced through chatbots? They often provide pertinent and contextualised answers to questions, with the ability to clarify and ask follow-up questions, in ways that were not possible before with (non-generative AI) natural language processing. LLMs are trained on and access data that exceeds what humans could retrieve. Moreover, they are flexible and can be applied to many different subjects, even though emerging research shows that narrow or specialised AI tools usually outperform general-purpose AI on the narrow tasks for which they were designed. In the case of education, this is for example the case for scoring assessment tasks. Finally, general-purpose GenAI tools can be turned into educational tools or combined with previous educational AI to enhance them with generative functionalities.

A series of well-known shortcomings are also specific to current GenAI, and are inherent to its technology, thus applying to some extent to all its derivatives. Because they are based on probabilistic models, they can “hallucinate”, that is, provide a wrong but plausible answer or invent some elements of the answer. They do not produce consistent results over time: asked to perform the same task several times, they will come up with different answers or productions, which is sometimes welcome, sometimes not. This may be because of regular system updates by providers, but mainly because of their probabilistic nature. And because they are trained on available datasets, their answers and other productions tend to reproduce the views and perspectives represented in those datasets, which are overwhelmingly based on English-speaking (and Western) cultures. For example, they may lack “representative accuracy” and fail to reflect the specific cultural and linguistic heritage of the local community in which they are used. As a result, their outputs require human supervision and scrutiny, and often more so than specialised, non-generative AI systems.

GenAI comes with a series of societal challenges that can also be highlighted. Large data centres on which AI depends, require large amounts of energy and water. One longer term impact of GenAI also lies in the inclusion and dissemination of AI-generated information and data in AI training datasets, which might either decrease the quality of future generated content or amplify some current limitations of our knowledge. This will make critical thinking and the development of metacognitive and higher-order thinking skills even more important than before. And as is the case for education, understanding is still emerging of how GenAI might transform societies and economies – as well as human cognition.

Opportunities and challenges for education

Like non-generative AI tools, the availability of GenAI tools creates both challenges and opportunities for education.

The ease with which students can use GenAI to perform assigned tasks is a challenge to education processes and systems, at least when students’ motivation to learn is faltering. The assumption that students’ work was evidence of learning and mastery does not necessarily hold when GenAI tools might have done it for them. For example, as automatic foreign language translators became remarkably more accurate in the past decades students were able to quickly perform traditional tasks of foreign language education such as translating to and from a foreign language with no or little associated learning. Similarly, the emergence of general-purpose GenAI tools could lead to a possible confusion between the successful performance of an educational task (an essay, for example) and the acquisition of the underlying content and procedural knowledge to accomplish this task. According to research on learning, learning happens when students actively (and often socially) work hard in a zone of proximal development, at the edge of what they already know and can do. Some uses of GenAI keep them out of that zone, rather than optimising how much time they spend in it. It will be important to minimise the risk of the use of GenAI tools undermining the acquisition of important knowledge, as well as complex and personal thinking abilities such as writing, analysing, interpreting, or appreciating the world. This also applies to teachers, education staffor policy makers when using GenAI tools to perform tasks whose value lies in the process as much as in the output of the task (e.g. an institution’s “strategic plan”).

GenAI tools also bring opportunities to the education sector, for improvement of teaching and learning as well as for the efficient and equitable administration of institutions and systems. One of the main benefits of AI in education lies in the personalisation of education, mainly for students, but also for teachers and administrators. In this context, personalisation means adapting learning tasks to individual needs, which can be fully balanced with collective education and the socialisation goal of education. In some cases, students can even perform adapted tasks while learning collectively. GenAI tools contribute to this agenda and can, for example, help students engage in a conversational way with intelligent tutoring systems that personalise educational tasks and feedback based on their current knowledge, level of engagement and other characteristics. Teachers can also benefit from GenAI tools to contextualise and adapt their lessons to their students’ interests or to receive feedback on their own teaching practices. GenAI could enable stakeholders to retrieve learning-analytics data by them asking questions in natural language, to explore the data further through follow-up questions that matter to them rather than the designers, and to obtain guidance on how such insights might inform action.

GenAI tools can also enhance education beyond personalisation. They allow for easier access to knowledge than search engines, by providing ready-made summaries, and even some level of analysis of the retrieved Internet information (which can include incorrect information). Educational GenAI offers the possibility to learn conversationally at any time of the day. In many cases, these tools are becoming a new layer added to non-generative educational AI tools, allowing for a more fluid natural language interface. GenAI also allows students to have educational experiences they could not have before, such as conversing with “digital personas” that roleplay different views and experiences or getting immediate feedback on their argumentation or on their writing style. Students learning in a second language or rural students can use GenAI tools for explanation or clarification when human support is not available.

In this context, education policy makers, institution leaders and other education stakeholders have to balance these opportunities and challenges. They may be tempted to prevent or delay as much as possible the use of GenAI tools so as to avoid the possible risks of GenAI on a precautionary basis. The unsupervised use of GenAI tools outside of school might still be difficult to prevent and does affect teaching, learning, assessment, and the comparability and fairness of education. Mastering the use of GenAI is also becoming a workforce development imperative. At some point in their educational journey, education systems will have to prepare students to use GenAI tools for professional purposes.

These Insights promote a second option for those willing to reflect on and steer the development, adoption and effective use of GenAI tools for educational purposes, in a curious, critical, responsible and disciplined manner. It not only assumes that GenAI tools will remain part of our societies, so that engaging with GenAI and exploring their constructive educational use (in balance with many other non-AI-related educational activities) will contribute to students’ preparation for life, including the labour market. It also recognises that despite the associated risks, if managed carefully and responsibly, GenAI tools provide many opportunities that students and teachers can leverage, especially if knowledge on the pedagogical use of general-purpose and educational tools is developed.

Seizing the opportunities of GenAI for education is a shared responsibility among all education stakeholders, from parents and families to teachers and students themselves. While the next section presents general guidance to seize GenAI opportunities in education, Table 14.2 highlights key pointers for all stakeholders.

GenAI Developers / Companies

Guidance to seize GenAI Opportunities

This section proposed elaborated guidance to support education stakeholders to use GenAI as a tool requiring human agency and serving human development. Rather than being driven by technical possibilities, novelty or economic opportunities, GenAI tools ought to be developed, used and assessed in education as a means to an educational end: serving pedagogical purposes, enriching meaningful learning, fostering equal opportunities, promoting the development of human knowledge, skills and wellbeing, and cultivating understanding, critical thinking, creativity and civic engagement. As discussed below, they are also expected to be privacy-preserving, transparent, able to explain their outputs, and, to the extent possible, working as effectively for all subpopulations of a country.

The Insights on the effective use of GenAI in education are organised in 5 sections:

1. Human-centred teaching and learning with GenAI;

2. Investment in GenAI-related educational R&D;

3. Enabling policy environment for trustworthy GenAI in education;

4. Effective GenAI infrastructure for all;

5. Human agency and human in the loop.