Generative AI is changing how students work, learn and find information. The public debate swings between two extremes. Some expect AI to revolutionise education. Others fear it is destroying learning. Science paints a more nuanced picture with no simple yes-or-no answer. The challenge is to understand that complexity and what it means for the responsible use of AI in education.
Existing research shows positive effects on average, but results vary considerably between studies, and it remains unclear which conditions favour positive and which favour negative effects. Depending on how they are used, AI tools can support learning or hinder it.
Digital technologies are already associated with stress and other strains, both in school and beyond. As generative AI becomes a more prominent part of students’ everyday lives, understanding its effects on well-being becomes increasingly important.
Generative AI systems can provide inaccurate or misleading answers, making it essential to assess their responses critically. At the same time, this creates an opportunity: learning to question AI-generated information may strengthen students’ critical thinking far beyond their use of AI.
GenAI studies the full chain from what AI systems can do, to how students use them, to what this means for learning and classroom practice. By bringing these dimensions together in a three-strand framework, the project can examine to what extent learning outcomes depend on the quality and capabilities of the AI system used. Measuring these capabilities also provides a reference point for interpreting the findings as future generations of AI become more advanced.
GenAI will develop a purpose-built test battery to assess at least ten leading generative AI systems across different STEM subjects and age groups. The resulting open benchmark will allow current and future AI systems to be evaluated and compared as technology evolves.
Assess subject-matter accuracy and performance.
Evaluate how well the systems explain concepts, recognise misconceptions and support learning processes.
Examine how effectively the systems adjust to different student learning levels.
At the heart of GenAI lies a cluster-randomised longitudinal intervention study involving around 8,000 students from eight countries, spanning two age groups and two STEM subject areas.
By comparing three carefully designed learning conditions, the study disentangles the effects of teaching materials, AI access and structured AI integration, generating robust evidence on when and how generative AI supports learning.
Teachers deliver the intervention in 3 separate classroom sessions, using comprehensively prepared, standardised teaching materials. Students are assessed repeatedly throughout the study, measuring STEM competences, AI literacy, scientific literacy, digital well-being and self-efficacy as learning develops over time.
During AI-supported lessons, interactions between students and AI are recorded. Together with repeated learning assessments, these data provide detailed insights into how students engage with generative AI during the intervention.
Semi-structured interviews with a total of 240 students and 40 teachers from eight countries take place before and after the intervention study
Capture how perceptions, expectations and experiences with AI develop over time.
Reflect on and validate the research findings together with teachers and educators.
Use teachers' assessment of what is feasible in everyday school life to shape practical recommendations.
To ensure our research output responds to real-world needs, we involve educators, policymakers and technology developers throughout the research process, drawing on their experience to shape resources that work in practice.
Link research with practice through regional roundtables with local authorities, educators and community representatives in all participating countries, European-level policy seminars and meetings with teacher educators. Together, they explore needs and opportunities for AI in education.
Give future users a direct role in reviewing and refining our interventions, findings and resources. Their feedback helps shape teaching materials, student guides and policy recommendations that are clear, relevant and suited to different educational settings.
The project creates a comprehensive, openly accessible knowledge base on the use of generative AI in STEM education, including validated tools, benchmarks, predictive models, policy recommendations, and educational resources.
Validated assessment instruments in eight languages, covering AI literacy, scientific literacy and digital wellbeing, among others
A test battery for evaluating generative AI systems, plus a continuously updated assessment of the most capable systems
Empirically grounded evidence on the conditions under which AI use supports learning and under which it does not
An empirical model linking AI capability to learning outcomes, allowing projections for future AI generations
Guidelines for teachers, a guide for students and policy recommendations, developed and validated together with practitioners
A cost-benefit analysis covering the financial and environmental resources of AI use in schools
More than ten scientific publications and professional development offers for teachers
There is no simple answer. Generation AI investigates what actually supports learning.
Whether you are a teacher, student, parent, policymaker or simply interested in how generative AI is changing education, we invite you to follow the project, discover opportunities to participate and engage with its findings.
Get in touch with us via email at anna.blattmann@ph-freiburg.de and follow us on Instagram, LinkedIn and Facebook.