Synthetic Consumer Insight Generation Using Large Language Models: Key Findings from New Research

On July 7, 2026, researchers Stephen L. France and Pia A. Albinsson published a study examining the use of large language models (LLMs) to generate synthetic consumer insight data. This research addresses the challenges of collecting consumer data, which can often be costly and time-consuming.

Understanding the Role of Large Language Models

Large language models (LLMs) have gained traction in various fields, including data-driven marketing. The study investigates whether these models can effectively simulate consumer responses for projective techniques, which are methods designed to uncover consumer associations and emotions. The authors tested LLM-generated responses across multiple tasks and compared them with human responses from a primary study focused on city tourism perceptions.

The findings indicate significant overlaps between human and LLM responses in broad topics and associations. However, the researchers also noted critical differences in linguistic structure and diversity generation. This suggests that while LLMs can mimic human-like responses, they do not fully replicate the complexity of human thought processes.

Key Findings on Response Quality and Diversity

The research emphasizes the importance of model and prompting choices in shaping the quality of responses generated by LLMs. By varying the prompting strategies and temperature settings, the researchers were able to influence the style and content of the responses. They recommend that marketers carefully consider these factors when utilizing LLMs for consumer data generation.

• Model choice: Different LLMs yield varying response qualities.
• Prompting strategies: Specific prompts can elicit more relevant and accurate consumer insights.
• Temperature settings: Adjusting this parameter affects response creativity and variability.

Limitations of LLM-Generated Data

Despite the promising results, the study acknowledges the limitations of using synthetic data generated by LLMs. The authors caution marketers to recognize these constraints, particularly in how LLMs may lack the nuanced understanding of human emotions and cultural contexts. They suggest further research to refine the techniques used in generating synthetic consumer data.

“While LLMs offer a novel approach to consumer insight generation, it is crucial to understand their limitations in replicating human emotional depth,” said France.