Abstract As generative AI becomes embedded in design studios, natural language increasingly mediates how design intent is specified, revised, and documented. This creates a pedagogical and assessment problem: final outputs may appear resolved while the criteria, constraints, assumptions, and decision paths behind them remain difficult to inspect. This paper examines successive GenAI-integrated iterations of an architectural design studio conducted over two years, using studio briefs and assessment rubrics as pedagogical instruments that structure what students are expected to state, record, and justify. From this reconstruction, the paper introduces Natural Language Designing (NLD), a conceptual and methodological framework for analysing how language functions as part of the design process in generative AI mediated workflows. NLD distinguishes three roles of language: representational, which specifies proposed artefacts; epistemic, which articulates criteria, constraints, assumptions, and evaluative tests; and meta-linguistic, which supports process control, documentation, traceability, and attribution. The analysis indicates that while prompts, workflow records, and visual outputs became increasingly central, the epistemic basis of design judgement often remained underspecified and therefore difficult to assess. The contribution is conceptual and methodological: the paper proposes a framework for reading and structuring studio tasks in terms of the linguistic evidence they require, with the broader aim of making design reasoning more inspectable in AI supported design education.
Nikolaeva et al. (Fri,) studied this question.