Elderspeak in Digital Health Interactions. A Comparative Analysis of Responses from Medical Professionals and AI Chatbots

The main objective of this study is to compare the responses of health professionals working in the Q&A services of an Italian multi-service medical platform with those provided by a chatbot, in relation to requests from users aged 65 or older. To this end, a corpus of approximately 12,000 question–answer sequences was collected using advanced web scraping techniques, from which 39 pairs of interactions with elderly users were manually selected and analysed. Each pair was labelled in order to investigate, on the one hand, the types of questions formulated by patients, together with their epistemic attitude and pragmatic function, and on the other, the degree of alignment, accommodation, and empathy conveyed by the responses of both healthcare professionals and the chatbot. The results show that elderly patients mainly resort to direct questions, especially wh-questions and polar questions, mainly aimed at requesting information or opinions. Healthcare professionals’ responses are generally pragmatically and epistemically aligned and adequately accommodated (i.e., attuned), although some under-accommodation (e.g. through the use of technical jargon) and over-accommodation (e.g. through excessive simplification) are observed; empathy, however, is limited. By contrast, chatbot responses, tested on a small sample of 3 users’ requests, are highly aligned and empathetic, but also mainly over-accommodated. The frequent use of bullet points, bold text, images, and claims of availability to simplify the content suggests the influence of age-related stereotypes, which assume diminished cognitive and digital skills among older users.Despite the limitations of the present study, its implications seem to be significant. The findings may help healthcare professionals engaged in Q&A services become more aware of their communicative styles and of their alignment with the needs of older adults, while also drawing attention to the risk of ageing-related stereotypes to which chatbots appear particularly vulnerable. More broadly, this knowledge can support practical recommendations for improving patient engagement and guide the development of design guidelines for chatbots.