Experimental Dataset on Recognition and Sharing of AI-Generated vs. Human-Generated Information

The datasets were derived from three online scenario-based behavioral experiments conducted via the Credamo platform. Participants were adult internet users from mainland China. Across the three studies, the experiments examined differences in information accuracy judgments and sharing intentions for content generated by humans versus artificial intelligence.Data collection took place between [4, 2025] and [11, 2025]. All tasks were completed online; therefore, no fine-grained spatial information was recorded. Basic demographic variables, including age, gender, and education level, were collected. Experimental materials consisted of both true and false information items generated either by human participants or by ChatGPT (GPT-4 model). In Study 3, an additional manipulation was introduced by varying whether the AI source was explicitly disclosed.All data were stored in SPSS (.sav) format. Each row in the dataset represents a participant’s evaluation of a single information item. Key variables include information type (true vs. false), information source (human vs. AI), AI source disclosure condition (present vs. absent), perceived accuracy ratings, online sharing intention ratings, offline sharing intention ratings, and demographic measures. Perceived accuracy and sharing intentions were assessed using 5-point Likert scales.Prior to analysis, the data were screened according to predefined criteria. Responses failing attention checks or showing abnormal response patterns were excluded. Missing data were minimal, and the exclusion of invalid cases did not substantively alter the main statistical results. Detailed data processing procedures and sample size information are reported in the Methods section of the manuscript.