Predictive Power for Product Sales: A Comparative Empirical Study of Consumer Reviews and Influencer Seeding Content

Consumer reviews and influencer posts provide two different views of the same product. This study asks whether either source improves product-sales prediction beyond product attributes. The sample contains 90 skincare products, 3,060 reviews, and 450 influencer posts from Tmall and Taobao. Four OLS models compare a product-control benchmark with review, influencer, and combined feature sets. Category-stratified five-fold cross-validation evaluates held-out products, while HC3 robust standard errors are used for inference. The control model performs best (adjusted R² = 0.651, pooled out-of-fold R² = 0.617, RMSE = 0.470). Adding content features slightly raises unadjusted R² but reduces adjusted and out-of-sample performance, and no focal content coefficient is statistically significant. In this small, platform-concentrated sample, product attributes contain most of the available predictive information. Visible review and influencer signals do not improve sales prediction once controls are included. Because the sales window, temporal order, promotional exposure, and brand strength are unavailable, the findings describe predictive associations rather than causal effects.