Metadiscourse in AI-generated hotel review summaries

Purpose In response to ever-increasing customers’ expectations for easily-accessible relevant information, Tripadvisor introduced an AI-powered feature that summarizes customer hotel reviews – AI Review Summaries (RSs). This study aims to explore how persuasion is manifested linguistically in these texts by focusing on metadiscourse features and their rhetorical effects. Design/methodology/approach This study is corpus-based and draws on Hyland’s (2005) interpersonal model of metadiscourse, slightly modified to cater for the specificities of the genre under investigation. RSs of 358 hotels in three most prominent global destinations were analyzed using a combination of top-down and bottom-up approaches, together with software-assisted and manual methods. Quantitative analysis is supplemented with functional interpretation. Findings The results point to the abundance of items that enhance the rational, credible and affective appeals, with slighter predominance of interactive over interactional categories and transitions and hedges being the most prevalent markers. These results directly reflect an RS overall purpose, on the one hand, and, on the other, could be seen as a reflection of their algorithmic nature. The analysis has also revealed the repetitive use of the same metadiscourse patterns coupled with the lexical non-diversity, which could be a potential hindrance to the intended persuasive effects. Originality/value This paper extends existing research on metadiscourse on AI-generated texts by examining texts from tourism domain. In doing so, it contributes to the study of AI hotel RSs by offering a discourse-oriented analytical perspective, thereby shedding light on the rhetorical nature of this emerging genre. Viewing metadiscourse as a component of persuasive rhetoric, the study shows how these resources contribute to the central purpose of the genre.