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

We see three practical reasons for why AI will be increasingly adopted by relevant stakeholders in institutional educational workflows.

First, content generation can yield strong increases in authoring efficiency that come with substantial economic cost reduction. For instance, Reza et al. (2025) found that PromptHive-an open-source collaborative prompt- authoring interface for the OATutor adaptive tutoring system-enabled subject-matter experts to produce AI- generated maths hints of comparable instructional quality to exclusively human-authored hints with no AI support. The tool also reduced perceived cognitive load by half, shortened authoring time by more than twenty-fold, and was found to be substantially more usable than the legacy authoring interface. In a controlled study with over 350 learners, the AI-assisted hints achieved student learning gains statistically indistinguishable from those of expert-written materials, demonstrating that human-centred prompt-engineering workflows can preserve expert control and quality standards while dramatically increasing the scalability of educational content creation. These improvements in content generation can be useful to both (a) higher education course instructors seeking to revise practice problems and assessment and (b) primary and secondary education instructors and content vendors seeking to revise curricular sequences.

Second, as many higher education institutions compete for the most capable and promising students, pressure to adopt AI that enables better course offerings and equivalences may increase. As transfer pathways improve, students will become more likely to select institutions that recognise a greater share of their prior learning and minimise credit loss, directly affecting enrolment and completion outcomes. Studies have found that seamless credit transfer is a strong predictor of degree completion among community college transfer students, and that improved articulation coverage can reduce time-to-degree and attrition (Monaghan and Attewell, 2015; Hodara et al., 2017). This is also a key development to make lifelong learning a reality.

Third, AI-augmented advising can directly improve other outcomes that education institutions care about-retention, on-time graduation, and advisor capacity-thereby improving cost effectiveness and institutional appeal. Advisor-facing recommender and triage tools can help tailor course plans, while course workload analytics enable programmes to audit sequences, rebalance hidden load, and reduce course withdrawal and failure where overload is detected (Pardos, Borchers and Yu, 2023; Borchers, Xu and Pardos, 2025). In parallel, machine-learning approaches to curriculum analytics can surface when nominally identical offerings fluctuate in difficulty over time and across groups, informing targeted redesign and quality assurance (Baucks, Schmucker and Wiskott, 2024; Baucks et al., 2024).

Across domains as varied as credit transfer, advising, admissions, and curriculum management, artificial intelligence is becoming part of the institutional infrastructure of higher education and beyond. These applications signal a shift in how institutions manage complexity, not by automating human decisions, but by introducing new forms of prediction, representation, and adaptive support. Properly designed, such tools augment institutional and education system judgment, surfacing course equivalencies that might otherwise be missed, helping advisors tailor guidance to individual students, and enabling administrators to detect inequities or inefficiencies across programmes.

With this integration comes a need for governance and privacy frameworks that can sustain trust. New privacy- preserving and open-source large language models, such as the Swiss-developed LLaMA-Open series, illustrate how innovation can proceed while respecting data sovereignty and transparency. Beyond regulatory frameworks, research can contribute frameworks for AI-supported analytics, as demonstrated in seminal models for responsible data use (Drachsler and Greller, 2016), offering templates for institutions developing AI policies today. Rapidly evolving technologies merit adaptive rather than static governance, that is, principles and review mechanisms that evolve alongside technology.

What distinguishes this emerging wave of institutional AI is less the technology itself than the collaborative ecosystems that enable it. Many of the most promising examples discussed in this chapter arose from partnerships among researchers, administrators, and platform developers. Such collaborations are critical for aligning technical design with institutional and end-user values and for empirically studying the consequences of AI adoption on efficiency, equity, and educational quality.

Although the focus of this chapter has been on higher education, many of the same institutional affordances extend naturally to lower levels of education as well as adult learning, where advising, assessment, and curriculum alignment face similar pressures. Generative models can assist teachers by creating or reviewing assessment items, tagging resources across standards, and generating formative feedback, thereby reducing workload and enhancing standardisation. It is likely that some institutional AI will merge toward a connected, data-informed educational ecosystem that links primary and secondary education and postsecondary systems, improving personalisation, equity, and mobility across educational pathways.