Predicting Consumer Behavior Using Large Language Models Augmented with User-Generated Content

As Large Language Models (LLMs) become increasingly embedded in marketing practice, their role has largely been descriptive or diagnostic: summarizing consumer sentiments, simulating surveys, and generating content. This paper asks a forward-looking question: Can LLMs, when augmented with user-generated content (UGC), predict what consumers will think or do next?

We argue that UGC-augmented LLMs can move beyond post-hoc analysis to forecast future consumer attitudes and behaviors. Across two studies, we test this proposition using different augmentation techniques — retrieval-augmented generation and fine-tuning — and prediction targets. In both experimental and empirical studies, we demonstrate the potentials of using UGC to adapt LLMs for predicting consumer attitudes and behaviors at both the group and individual level.