AI Models Might Be Able to Predict What You'll Buy Better Than You Can
Researchers demonstrated that large language models (LLMs) can reliably predict human purchase intent, surpassing traditional marketing tools. By introducing the "Semantic Similarity Rating" (SSR) method, they converted LLMs' natural language responses into structured Likert-scale ratings, matching human survey results about personal-care products. This process mapped open-ended responses to reference statements in embedding space, yielding distributions that closely mirrored real consumer data—achieving 90% of human response consistency. The approach suggests that LLMs encode attitudinal reasoning, enabling large-scale, low-cost product testing and market research. However, limitations remain: the current method relies on narrow product categories, reference-dependent phrasing, and potentially biased human survey data. Critics note that mapping language embeddings to attitudes may not capture nuance, and the approach’s reliability, while high, is imperfect. Ethical risks include potential misuse for persuasion or psychological manipulation. While promising, SSR’s generalizability to other domains and its representation of true human diversity remain unsettled, signaling opportunities and challenges as LLM-based synthetic consumer modeling advances.
