Response Corpus and Coding Data for "Evaluating Sports Nutrition Advice Provided by General-Purpose Large Language Models for Recreational Female Marathon Runners: A Multidimensional Descriptive Audit"

This dataset supports the study “Evaluating Sports Nutrition Advice Provided by General-Purpose Large Language Models for Recreational Female Marathon Runners: A Multidimensional Descriptive Audit.” The study evaluated 750 responses generated by five general-purpose large language model configurations using 15 standardized consumer-style prompts, with 10 independent runs per prompt and model configuration. The deposited materials include the complete 750-response raw corpus and the paired response-level coding sheets used for inter-rater reproducibility assessment. The coding dataset contains blinded response identifiers, response-level indicator coding, and agreement statistics used to assess raw agreement, Cohen’s kappa, exact agreement for numerical extraction fields, and ICC(2,1). These materials are provided to support transparency, reproducibility, and independent verification of the descriptive findings reported in the manuscript.File access is restricted because the raw response corpus contains AI-generated outputs from third-party services that remain subject to the applicable service-provider terms. Access may be provided for scholarly verification and reproducibility purposes subject to those applicable terms.