The Impact of Recommender Systems on Content Consumption and Production: Evidence from Field Experiments and Structural Modeling

Abstract

Online content-sharing platforms such as Facebook and TikTok, where users act as both content creators and consumers, have become integral to daily life, relying on complex algorithms to recommend user-generated content (UGC). While prior research and industry efforts have primarily focused on designing recommender systems to encourage users to consume more content, the impact of the content recommended to users on their own content production remains understudied. To address this gap, we conducted a large-scale randomized field experiment on a major video-sharing platform. Users were randomly assigned to either a treatment or control condition, and we manipulated content recommendations by excluding a subset of highly popular creators from being recommended to treated users. This reduced recommendation quality of treated users compared to control counterparts. Our results indicate that this decline led to a 1.34% decrease in video-watching time but a 2.71% increase in videos uploaded by treated users. These findings reveal a trade-off in recommender system design: while higher recommendation quality boosts consumption, it may suppress production. To optimize recommendations, we developed a structural model in which users’ choices between consuming and producing content are inversely affected by recommendation quality. Counterfactual analyses suggest that slightly lowering recommendation quality—rather than prioritizing the most popular creators—can maximize the overall value of both production and consumption for platforms. This highlights the need to balance content consumption and production when designing recommender systems on these social media platforms.

Keywords: User-Generated Content, Recommender System, Field Experiment, Structural Model, Platform Operations, Production