What shapes user engagement with bot-like accounts on social Media? Evidence from Twitter

Users on social media platforms increasingly encounter and interact with automated accounts. While prior research has mainly examined how social bots operate and shape online environments, less is known about how human users contribute to human-bot interaction through their own engagement behavior. Drawing on 33,066 observed human-bot interaction ties from the TwiBot-22 Twitter/X dataset, this study adopts a human-centered perspective to examine human users’ engagement strength with social bot accounts. Grounded in the Computers Are Social Actors framework and social psychological theories, we investigate how bots’ following behavior and tweet-content similarity relate to human users’ engagement and whether these associations vary with Botometer-inferred human-likeness, an algorithmic measure conceptually related to perceived humanness. The results show that content similarity is positively associated with human users’ engagement strength, and this association is more pronounced at higher levels of Botometer-inferred human-likeness. Supplementary analyses further identify positive associations for both topical and sentiment alignment. In contrast, following behavior from bots generally shows a negative association. These findings contribute to research on social media interaction by highlighting how social cues and human-like characteristics jointly relate to human users’ engagement with social bots. They also provide insights for platform governance and the design of emerging AI-mediated service agents.