Algorithmic accommodation: linguistic alignment in human-AI relational engagement

Abstract Communication accommodation is critical to relationship development, yet how verbal mimicry and coordination operates in the relational engagement of human-AI intimacy remains incompletely understood. Using a corpus of over 11,000 conversation snippets shared by more than 5,000 users of Replika – a widely adopted AI companion application, this study examines how linguistic alignment between users and AI companions relate to the intensity and depth of user engagement in real-world development of human-AI relationships. Leveraging computational analyses of conversation-level syntactic and semantic alignment, results from linear and logistic regression models indicate that meaning-level (i.e., semantic) and structure-level (i.e., syntactic) alignment were associated with user’s relational engagement in distinct ways. Specifically, higher semantic alignment was linked to greater interaction intensity, whereas higher syntactic alignment was associated with deeper self-disclosures. These findings suggest that semantic and syntactic alignment index distinct verbal coordination processes in human-AI communication and highlight the need to reconsider how communication accommodation theory applies and adapts to human–machine communication, given the algorithmic nature of AI interlocutors.