LLMs Generate Synthetic Consumer Insights for Marketing Research.

Modern marketing heavily relies on extensive consumer data to understand preferences and behaviors. However, collecting this data can be expensive, time-consuming, and difficult to scale effectively. This study explores the potential of Large Language Models (LLMs) to generate synthetic consumer data, specifically for projective techniques aimed at uncovering consumer associations, emotions, wants, and needs. The researchers tested LLM-generated responses across various projective tasks, using different LLMs, prompting strategies, and temperature settings. These synthetic responses were then compared against human responses obtained from a primary research study focusing on perceptions of city tourism destinations. Analysis involved linguistic measures, diversity metrics, topic models, and top-term analyses. The findings indicate a substantial overlap between human and LLM responses in terms of broad topics and associations. However, notable differences were observed in stylistic elements, linguistic structure, and how diversity in responses was generated. The study provides recommendations for optimizing LLM utilization in synthetic data generation, emphasizing the impact of model and prompt choices on response quality, while also highlighting the inherent limitations of using LLMs for this purpose.