Visibility of AI-Generated Short Dramas in Algorithmic Recommendation Feeds: Evidence from a Dual-Account Experiment

Abstract

In recent years, generative AI has given rise to AI-generated short dramas (AIGSDs), which have gained massive popularity on short-video platforms such as TikTok. However, it remains unclear whether the popularity of such content is mainly driven by active user demand or algorithmic supply. This study investigates the relative roles of user demand and algorithmic supply in shaping the visibility of AIGSDs and adopts a mixed-methods design, combining a 14-day controlled dual-account experiment involving four simulated user profiles (active search vs. passive browsing × male vs. female) with in-depth interviews conducted before and after the experiment (N = 4). The results showed that active search significantly increased the penetration rate of AI-generated short dramas (25.3% vs. 11.8%, p < 0.001), and the amplification effect persisted for at least three days. Emotionally compensatory and curiosity-driven strange content received higher exposure than conventional plot-driven content. Notably, a divergence between behavioral engagement and attitudinal preference was observed: high completion rates (83.6%) coexisted with low genuine preference (22.5%) and high curiosity-driven viewing (76.4%). Gender differences mainly appeared in content preferences rather tha n in overall penetration rates. These findings reveal a dual-construction mechanism underlying the popularity of AI-generated short dramas, whereby user demand and algorithmic supply interact to shape content consumption. This study provides empirical evidence for filter bubble theory, offers a critical perspective on the optimization logic of algorithmic platforms, and generates practical implications for algorithm governance and media literacy education.