This study investigates a fundamental question in AI-human interaction: To what extent do language models internalize and propagate human opinions when forming their own judgments? Using business reviews as a microcosm of human opinion expression, we examine how LLMs sequentially update their evaluations based on human-generated content. Unlike traditional sentiment analysis that treats reviews independently, our framework models opinion formation as a social learning process where LLMs act as synthetic agents influenced by the expressed opinions of human reviewers. We reveal that LLMs not only analyze content but also calibrate to collective human sentiment, raising important questions about AI autonomy, social influence dynamics, and the construction of "artificial opinion" in review systems.