Give and Take: Perceptions of a Conversational Coach Agent in Fitness Trackers MHCI014

While Personal Informatics (PI) tools utilise data visualisations to communicate behaviour, users often struggle to make sense of their data and translate it to actionable insights. Conversational Agents (CAs) offer potential for improved access to PI data, yet their role in PI tools remains under-explored. We conducted a two-week user study with journals, interviews and logging with n = 36 participants using a novel commercial fitness tracker with an embedded CA. We identified the give and take principle as essential for meaningful sensemaking with a CA—a dynamic resulting in more effective interactions given users’ inputs ( give ) are met with prompts that are sufficiently specific and built upon prior data engagement ( take ). A critical point was how users perceived the CA during their initial interactions, with first impressions often determining further engagement. We contribute insights into how CAs can support or hinder the PI experience, offering implications for future PI system designs.