EVA's Mapping of Assessment Practice in Danish Upper-Secondary Schools — How 12 Gymnasiums Grade Work in the Age of Generative AI

In March 2025 the Danish Evaluation Institute (EVA), on commission from the Danish Agency for Education and Quality (STUK), published a mapping of assessment practice in upper-secondary education (stx, hhx, htx, hf) in the light of generative AI. Because generative AI is generally banned in gymnasium exams, little is known about how to assess work where it is used, so the mapping draws on interviews with leaders and teachers at 12 gymnasiums chosen for their focus on integrating generative AI, interviews with representatives of 7 universities that have developed AI guidelines for selected exams, a literature review of Danish and international studies, and a three-expert sounding board.
The report finds that generative AI mainly challenges the assessment of students' independent proficiency in written work. Teachers spend considerable time and energy checking for cheating, and report that diligent students may even insert errors on purpose for fear of being wrongly accused. The literature review adds that assessors may mark more harshly when they merely suspect AI use.
Teachers respond by turning up familiar techniques: (1) exercises without aids, to assess basic subject proficiency; (2) a focus on students' work process in projects and open tasks, including meta-skills such as critical thinking and reflection; (3) oral presentations or defences based on written work; and (4) attention to students' interaction with AI, for example by attaching their prompts and a reflection on how they used them. For exams, the report points to two paths: written on-site exams that include unaided elements, and written formats with an oral defence where digital aids are permitted. It also stresses that teachers need competence both to teach learning-supportive use of AI and to take that use into account when grading.
This is a qualitative mapping, not an impact evaluation: it reports no learning outcomes, and its sample was purposively selected from schools already engaged with generative AI, so findings may not generalise to all gymnasiums.

Read the full analysis: https://www.eva.dk/ungdomsuddannelse/gymnasiale-uddannelser/undervisning-paa-gymnasiale-uddannelser/ai-i-undervisningen