When AI Only Asks: How Question-Driven Dialogue Shapes Prewriting in the Classroom

Anne Katrine Westbye¹Harald Eriksen¹Marte Blikstad-Balas^2*

• ¹OsloMet - storbyuniversitetet, Oslo, Norway
• ²Universitetet i Oslo utdanningsvitenskapelige fakultet, Oslo, Norway

Generative artificial intelligence has altered the conditions under which pupils learn to write, raising concerns that the slow, effortful work of idea generation and elaboration may be bypassed when fluent text can be produced instantly. This small-scale exploratory study examines how a deliberately constrained chatbot can support pupils' thinking during prewriting by sustaining dialogue rather than generating text. Seventeen lower-secondary pupils in Norway interacted with a GPT-4–based chatbot programmed to ask one question at a time while refusing to generate text on pupils' behalf as they prepared to write about their hometown. Using reflexive content analysis, we examined 201 question–response pairs across entire prewriting dialogues to trace how pupils took up, reshaped, limited, or resisted the chatbot's questions over time. Drawing on both a cognitive and a sociocultural view of writing, the analysis shows that sensory and memory questions often supported elaboration by anchoring attention in concrete experience, while reflective questions were productive mainly when grounded in such detail. Pupils engaged with the question-driven format in markedly different ways: some sustained stable descriptive engagement; others negotiated relevance through redirection, and some constrained the dialogue through minimal uptake or repeated resistance. The findings suggest that the value of a question-only chatbot lies not in the questions themselves, but in how pupils engage with the dialogic space such constraints create. Implications for practice is that constrained generative AI can support students in the prewriting phase of classroom writing, and that some students need additional support and motivation from the teacher or peer in such sessions