Digitalisation and the rapid development of artificial intelligence (AI) have created new opportunities and challenges for education. The Swedish National Agency for Education emphasizes that schools are responsible for developing students' digital competence and their ability to use digital tools critically and responsibly (Skolverket, 2024). Previous research and reports from the Swedish Schools Inspectorate further highlight that the educational value of digital tools depends on how they are integrated into teaching and support students' learning processes (Skolinspektionen, 2019). 

This study examines how a structured teaching approach involving generative artificial intelligence (AI), AI-support sheets and teacher modelling can be related to students' conceptual understanding in physics education in grade 6. The study was conducted as a practice-based intervention in a Swedish primary school classroom where students used generative AI tools, such as ChatGPT and Google Gemini, during a structured teaching unit on the universe. The aim was to investigate how this teaching approach could be related to students' conceptual understanding and use of subject-specific concepts. The study is grounded in a sociocultural perspective on learning, where concepts such as mediation, scaffolding and linguistic tools functioned as analytical concepts (Säljö, 2014). The empirical material consisted of pre- and post-tests focusing on central physics concepts, such as gravity, orbit, galaxy and light-year. The analysis combined quantitative and qualitative approaches to examine both measurable changes in students' results and the development of their use of scientific concepts (Bryman, 2018). The findings indicate that several students demonstrated a more developed use of scientific concepts and more elaborated explanations in the post-test. The results also suggest that generative AI, when combined with AI-support sheets and teacher modelling, may function as linguistic and pedagogical support by promoting reflection, comparison and conceptual reasoning. At the same time, the study highlights challenges related to source criticism, reliability and the extent to which students' responses reflected their own understanding or AI-generated formulations. The study contributes with practice-based knowledge about opportunities and limitations related to the use of AI in science education.