Anticipated Acceptability of AI Health Coaches

The rapid development of artificial intelligence and connected health technologies positions virtual nutrition coaches (VNCs) as promising tools for supporting dietary behaviour change and addressing major public health challenges related to inadequate nutrition and physical inactivity (Remington et al., 2010). While the literature provides strict guidelines regarding usability, personalization, and behaviour change techniques (Sharpe et al., 2017; König et al., 2021), technology adoption begins prior to use. Individuals construct expectations by mentally simulating their future interactions with the device, a process conceptualized as anticipated interaction mental simulation (Taylor & Schneider, 1989; Hakimi & Valette-Florence, 2023). Yet the ways in which non-users develop these representations remain underexplored. This study investigates the formation of anticipated acceptability of VNCs among individuals who have never used this technology. A qualitative approach was conducted through an online community facilitated over a two-week period (Helme-Guizon, 2024). Of the forty individuals recruited, twenty-one participated actively. Thematic content analysis (Bardin, 2013) identified five interdependent themes: relationship with food, barriers to change, functional expectations, human–machine interaction, and ethical and social considerations. The findings indicate that non-users engage in sophisticated mental simulations, assessing the coach's usefulness, reliability, and practicality while anticipating affective dynamics such as support, judgment, or loss of autonomy (Hakimi & Valette-Florence, 2023). However, these simulations are not solely guided by technological considerations; they are deeply rooted in individuals' relationship with food, which acts as a structuring filter expectations and resistance. Health-oriented participants tend to envision the VNC as an optimization tool, whereas those attached to pleasure express greater reservations toward features perceived as normative. Nevertheless, a large proportion project the coach as a flexible mediator, capable of supporting everyday trade-offs rather than imposing strict control. This expectation of mediation constitutes a central contribution, standing in tension with dominant models grounded in self-monitoring and behavioural quantification (König et al., 2021). Participants show limited interest in intensive nutritional tracking and anticipate the psychological risks associated with excessive quantification (Lyzwinski et al., 2017). Perceived usefulness therefore relies less on knowledge acquisition than on the technology's ability to reduce the gap between knowing and doing, positioning the VNC as an operational assistant rather than a pedagogical authority. The results also reveal a structural tension within human–machine interaction: although empathy and personalization are expected, increased anthropomorphism does not systematically enhance acceptability, thereby nuancing prior work on embodiment and computational empathy (Cassell et al., 2000; Lisetti et al., 2013). Participants seek a dynamic balance, a coach sufficiently humanized to understand emotional experiences, yet sufficiently distant to preserve autonomy. Finally, concerns related to data protection, scientific credibility, and the social embeddedness of eating practices emerge as moderators of overall acceptability. Overall, this research reframes anticipated acceptability as a relational, identity-based, and contextual phenomenon rather than a strictly functional one. By focusing on non-users, it uncovers latent expectations and resistance mechanisms that remain largely invisible in post-adoption studies and calls for moving beyond prescriptive logics to design adaptive virtual nutrition coaches aligned with the psychological and social realities of dietary practices.