Generative AI now allows architects to converse with simulated users who speak in the first person about spatial needs, discomforts and preferences. This creates a new challenge for empathetic and inclusive design: synthetic experience can sound like lived testimony and then migrate into briefs, programmes and spatial requirements with increasing authority. This paper addresses how AI-generated experiential language changes epistemic status as it is translated into architectural decision-making. Using critical conceptual analysis and a comparative reading of two published AI-assisted spatial-design cases, the paper develops the concept of testimonial-to-requirement transformation. It identifies four recurrent shifts: loss of source attribution, hardening of epistemic modality, expansion of experiential scope and borrowed authority through formalisation. The paper concludes that empathy in design cannot be reduced to plausible simulation. Inclusive practice requires provenance, explicit uncertainty and opportunities for contestation before synthetic claims are treated as user knowledge.