This presentation is based on a larger research project about novel and film summaries on Wikipedia where a recurrent dissensus as to which word summarizes best what happens in a fictional sex scene (typically seduction vs rape) was spotted by exporting and searching revision histories. The linguistic consistency and recurrence of such words shifts suggests these findings cannot be explained solely by interindividual subjective variation in interpretation and chance : can we predict whether a reader / contributor will make this type of discursive shift on the basis of their social identity, such as their gender ? A methodological difficulty is that identity online and particularly on Wikipedia is often blurred and ranges from full presentation to a mere IP address, in a mostly-male community of contributors. Cues about the contributor's gender in metadatas and user pages (name, userboxes) are the exception. However, by studying the users' history of contributions to other articles (including unregistered contributors identified through their IP address), we could trace patterns rooted in politicized practices of verbal hygiene rather than gender identity. Further investigation suggests that some contributors care about sexual violence vocabulary in fiction because they care and learnt to care about the political meaning of language in general. Three main discursive patterns appear : 1/ contributors make identical discursive shifts (renaming an action "rape") in various fictional narratives ; 2/ contributors specialize in articles about sexual violence, whether fictional or real, and feminism-related themes ; 3/ contributors display recurrent verbal hygiene practices linked to progressivism (anti-racism, transgender rights). Although our data does not allow a quantitative approach that could generalize these findings, the qualitative focus on a sample of contributor profiles still provides a more robust hypothesis about discursive behaviours regarding gender-related subjects online.