Does transparency matter when interacting with social bots? An experiment on user engagement behaviors in Chinese social media

Purpose While transparency is widely promoted as a principle of responsible AI design, its behavioral consequences remain contested, particularly in Chinese social media contexts. This study examines whether different levels of social bot transparency reduce or facilitate user engagement, and whether these effects depend on users' prior knowledge of social bots. Design/methodology/approach An online experiment was conducted in a simulated Weibo environment. A total of 731 participants were randomly assigned to one of three transparency conditions: low, medium, or high transparency. User engagement behaviors were measured through participants' behavioral choices, including reading, liking/sharing, replying, and combined engagement actions. Multinomial logistic regression and subgroup analyses were used to examine the effects of transparency and the moderating role of prior knowledge. Findings Social bot transparency significantly reduces active user engagement. Compared with high transparency, low transparency increases the likelihood of liking/sharing, replying, and combined reply-based engagement rather than merely reading. Medium transparency produces engagement patterns similar to high transparency, suggesting that even ambiguous bot identity cues may activate non-human attribution and suppress active engagement. Prior knowledge moderates this relationship: users with limited prior knowledge are more sensitive to transparency cues, whereas those with higher prior knowledge show more stable engagement across transparency conditions. Originality/value This study extends social bot transparency research beyond binary disclosure by identifying a categorical identity-attribution mechanism. It also clarifies prior knowledge as a boundary condition of algorithm aversion in human-AI interaction.