Sociopragmatic variation in AI-generated turkish requests

This study aimed to examine how ChatGPT, Gemini, and Microsoft Copilot realize Turkish requests across systematically structured conditions of relative power, social distance, and degree of imposition. Twelve request scenarios were separately presented to each model five times, yielding 180 outputs. The responses were analyzed using a CCSARP-informed coding framework focusing on request strategy, second-person marking, and interpersonal and mitigating resources. All 180 outputs employed conventionally indirect Query Preparatory strategies, indicating complete convergence in the realization of the request head act. Greater variation emerged at the sociopragmatic level. Social distance showed the most consistent pattern in second-person marking, although this pattern differed according to relative-power configuration. Gemini used the broadest range of interpersonal resources, ChatGPT showed more selective elaboration, and Microsoft Copilot showed the least interpersonal elaboration within the coded categories. Repeated generations showed complete stability in request strategy and high stability in second-person marking. The findings show that pragmalinguistic convergence can coexist with model-specific sociopragmatic variation in AI-generated Turkish requests.