Abstract

Purpose: This study aims to explore the potential of virtual data by examining whether AI-generated data using ChatGPT shows statistical differences compared to real survey data. Research design, data and methodology: Based on the existing local food purchase motivation model, first- and second-round virtual datasets were generated using ChatGPT. In the second dataset, demographic characteristics, social desirability bias, and midpoint response ratios were adjusted to minimize structural differences from the real data. Results: The analysis revealed that the second virtual dataset produced path coefficients and statistical significance similar to those of the real data in major relationships, such as the effect of brand equity on consumer attitude and the influence of consumer attitude on purchase motivation. Differences appeared only in some relationships, specifically in the path from brand equity to perceived value and from perceived value to consumer attitude. Conclusions: This study empirically confirmed that virtual data can structurally converge with real data. Academically, it contributes to bias control and reproducibility, while practically, it highlights the value of virtual data as a decision-support tool that enables targeted simulations of marketing strategies and predictions of consumer responses.