Dialogic Artificial Intelligence in the Architectural Design Studio: An Empirical Study of the AI Sparring Model

Integrating generative artificial intelligence into architectural education raises a question not only of whether AI helps, but of how its use should be structured. This paper empirically examines the AI Sparring model—a dialogic, critical approach to student–AI interaction—through a workshop in which architecture students designed a pavilion in a fortress. The three self-selected groups differed in the role AI played in the formation of the concept, not in its final presentation: Group A developed concepts traditionally, through sketches and physical working models, with AI serving only as a rendering tool at the end; Group B used AI directly as a generative tool; and Group C worked through structured critical dialogue with AI. The works were independently rated by three evaluators from three countries, alongside keyword analysis and assessment of reflective depth and theoretical references. Group C produced the most consistent concept quality, with no weak works, yet the single highest-rated work came from the traditional group—Sparring raises the “floor” rather than the “ceiling” of quality. At the same time, Group C reported the lowest sense of authorship, and qualitative analysis revealed “simulated depth”: theoretical references cited without genuine engagement. Because of the small, self-selected samples, the findings are read as exploratory and hypothesis-generating. The paper sheds light on the limits of reflective AI pedagogy and proposes concrete revisions to the model.