Abstract
•Patient interviews and surveys show how chatbots shape health rapport.•The GAAFFE framework captures patient perceptions of human–chatbot rapport.•Anthropomorphism is a moderator of all dimensions of rapport.•A taxonomy of human–chatbot rapport supports research and conversation design.
Chatbots are becoming ubiquitous for everyday tasks, such as getting customer support, banking or personal assistance. Health services are also increasingly using chatbot applications, for example to facilitate appointment booking. This paper investigates user perceptions of rapport with a generative AI chatbot which invites patients to and makes bookings for cervical cancer screening. Using seven interviews with patients at the trial site and a survey of patients eligible for cervical screening via the NHS (n = 300), analysed through content analysis based on the GAAFFE framework, we investigate how patients perceive the way this chatbot, named Asa, manages rapport and what positive or negative perceptions of rapport patients express. With a particular focus on patients’ reactions to the language and interaction design in Asa’s dialogue flow, we show that human-like features act as moderators across all dimensions of the rapport-management framework. We also show that rapport perceptions are influenced by other factors, e.g. the messaging channel. Based on these insights, we present a taxonomy of human-chatbot rapport for text-based conversational AI applications and show how this taxonomy and our findings can be used by conversation designers and researchers. The paper makes a theoretical contribution by applying the GAAFFE framework to human-AI interaction, paving the way for training and evaluating chatbots through the lens of rapport. It also provides theoretical underpinning for the ‘Machines are Social Actors’ framework through rapport theory.