Auditing cultural capability in generative AI: A persona-based evaluation of Chinese LLMs in higher education

This dataset supports the study “Auditing cultural capability in generative AI: A persona-based evaluation of Chinese LLMs in higher education.” The study examines how three Chinese large language models, DeepSeek, Kimi, and Qwen, perform cultural capability when simulating higher education personas. The dataset includes persona prompts, model-generated responses, Cultural Intelligence questionnaire materials, CQ scoring data, analysis scripts, and generated figures. The data can be used to verify the tables and figures reported in the article and to examine differences in model performance across educational roles and regional identities.