Can AI convincingly answer consumer surveys? This project compares verified human responses with AI-generated responses, with and without respondent characteristics, to examine how closely large language models can mimic consumer answering behaviour and whether such responses can be detected

Online surveys are widely used in sensory and consumer research to understand consumer perceptions, attitudes, and preferences. However, recent advances in AI, particularly large language models, have introduced a growing risk that survey responses may be generated or heavily assisted by automated systems rather than by real participants. Recent research suggests that large language models can produce plausible survey answers, but that such responses may differ from human data in important ways, including reduced variability, unstable bias patterns, and sensitivity to prompting (Bisbee et al., 2024; Boelaert et al., 2025; Jansen et al., 2023). At the same time, emerging work shows that participants already use generative AI in open-ended survey settings and that AI-generated responses can be difficult to detect with standard quality checks (Lebrun et al., 2024; Zhang et al., 2025).

This master’s project aims to examine how responses generated by state-of-the-art large language models compare to verified human responses in consumer survey settings. The focus will be on differences in answering behaviour, response patterns, and language use across both rating-scale questions and open-ended responses. The project will also examine whether prompting AI with predefined respondent characteristics makes its responses appear more human-like, or whether important differences remain.

The student will:

• Generate consumer survey responses using selected large language models, both with and without predefined respondent characteristics
• Compare AI generated responses to human responses in terms of response behaviour and consistency such as rating scales and open-ended survey questions, and across food consumer contexts (e.g., different product categories, sensory attributes, hedonic vs. attribute-based questions)
• Investigate whether providing AI with respondent characteristics affects how it answers survey questions
• Examine differences in language use and content between human and AI responses to open-ended questions
• Explore whether AI generated responses can be distinguished from human responses based on typical answering patterns in consumer surveys

The project contributes to current work on survey integrity and data quality by testing these questions in a food consumer research context. The aim is not only to assess how closely AI can mimic human respondents, but also to identify practical indicators that may help researchers recognize synthetic or AI-assisted survey data.

The project relates to SDG 9 (Industry, Innovation and Infrastructure), through its focus on AI-driven research methods, and SDG 12 (Responsible Consumption and Production), through its food consumer research context.


Reference list

• Bisbee, J., Clinton, J. D., Dorff, C., Kenkel, B., & Larson, J. M. (2024). Synthetic replacements for human survey data? The perils of large language models. Political Analysis, 32(4), 401–416. https://doi.org/10.1017/pan.2024.5
• Boelaert, J., Coavoux, S., Ollion, É., Petev, I., & Präg, P. (2025). Machine bias. How do generative language models answer opinion polls? Sociological Methods & Research, 54(3). https://doi.org/10.1177/00491241251330582
• Jansen, B. J., Jung, S.-g., & Salminen, J. (2023). Employing large language models in survey research. Natural Language Processing Journal, 4, 100020. https://doi.org/10.1016/j.nlp.2023.100020
• Lebrun, B., Temtsin, S., Vonasch, A. J., & Bartneck, C. (2024). Detecting the corruption of online questionnaires by artificial intelligence. Frontiers in Robotics and AI, 10, 1277635. https://doi.org/10.3389/frobt.2023.1277635
• Zhang, S., Xu, J., & Alvero, A. J. (2025). Generative AI meets open-ended survey responses: Research participant use of AI and homogenization. Sociological Methods & Research, 54(3). https://doi.org/10.1177/00491241251327130