Understanding what good education is: a phenomenographic investigation of university teachers' understandings of AI

Synthesis: Understanding what good education is — a phenomenographic study interviewing 16 teachers in Danish higher education to map how they understand and frame AI in teaching and student learning. It identifies three qualitatively different ways of understanding the relationship between teaching and AI, three ways of understanding the relationship between learning and AI, and four approaches to designing for student learning experiences around AI — highlighting the plurality and messiness of educational practice, and the need to attend to agency and teachers' labor.

Stenalt (2026) shifts focus from students and tools to teachers' conceptions of AI — what teachers believe counts as good education in an AI-integrated context. Rather than prescribing how AI should be used, it surfaces the variation in how educators themselves conceptualize AI's role, which shapes how they design AI into learning experiences.

Method

• Approach: Phenomenography (a qualitative method for mapping the qualitatively different ways people experience/conceive a phenomenon).
• Sample: 16 teachers in Danish higher education.
• Focus: Teachers' conceptions of AI in higher-education teaching processes and student learning processes, plus their approaches to designing AI into student learning experiences.

Key Findings

• Three conceptions of teaching–AI relationship: qualitatively different ways teachers understand the relationship between teaching and AI.
• Three conceptions of learning–AI relationship: qualitatively different ways of understanding AI in relation to student learning.
• Four design approaches: four qualitatively different approaches to designing AI into student learning experiences.
• Plurality and messiness: the outcome spaces highlight that educational practice is plural and messy — there is no single "right" way to understand AI in education.
• Agency and teachers' labor: the study points to the need to attend to teacher agency and the labor of integrating AI.

What this means for practice

• Instructors. Name the conception of AI your course design assumes — correct-tool knowledge, a resource for individual sense-making, or a disciplinary collaborator — because the four design approaches this study identified assign students different roles, from recipients of information to collaborators.
• Instructors. Build tasks that require judgment about when not to use AI: the interviewees overwhelmingly framed AI Literacy as "how to do it" and rarely voiced opting out of AI activities or prioritizing differently.
• Instructors. Use the four design approaches as shared vocabulary in course teams, so that design discussions surface which elements (purpose, context, values, method, technology) are being prioritized rather than assuming one correct use of AI.
• Faculty developers. Design professional development around teachers' existing conceptions instead of tool training: the 16 interviewees held three conceptions of teaching-with-AI and three of learning-with-AI, so a single workshop model reaches only part of any faculty group.
• Administrators. Treat AI-related concern as a considered position rather than a competence deficit: these teachers, positioned as actively developing their practice, did not adopt AI uncritically, and the study ties teacher agency and the labor of integration to how AI actually enters teaching.

Limitations

• 16 teachers from 2 of Denmark's 8 universities, recruited by snowballing through university teachers and academic developers who were already adopting AI in teaching — the outcome spaces therefore over-represent engaged adopters, and the authors state that including teachers with less interest in AI would allow more extensive exploration.
• The evidence is single semi-structured interviews of 45–60 minutes each; participants' course details and profiles were self-reported during the interviews rather than verified against institutional records.
• Conceptions and approaches are interpreted as situated responses to context, and the authors state the analysis does not provide in-depth insight into teachers' agentic will and power — the study maps variation, not its causes.
• No longitudinal component: the authors call for further investigation of how the identified positions develop and whether they remain robust over time.

Citation

Stenalt, M. H. (2026). Understanding what good education is: a phenomenographic investigation of university teachers' understandings of AI. Teaching in Higher Education.