Intervention science faces a hazardous paradox: on the one hand, vulnerable populations (e.g., patients, people from low socio-economic background, older adults) are those for whom adoption of healthy behaviours is most urgent; on the other hand, behaviour change models are less predictive, and interventions less successful, in these populations. This commentary presents four reasons that may explain this issue: (1) research mostly focuses on what causes behaviour and how to change it, at the expense of investigating among whom and under what conditions models are valid; (2) models put an undue emphasis on individual cognitions; (3) most studies are not conducted on vulnerable populations; (4) most researchers are from high-income countries. Several avenues are proposed to address this issue: (a) providing a central place to the context and audience in health behaviour change modelisation, through collaborations with researchers from other disciplines and countries, and with members of the targeted audience; (b) better reporting samples’ socio-demographic characteristics and increasing samples’ diversity; (c) using more rigorous and innovative designs (e.g., powered randomised controlled trials, N-of-1 trials, intensive longitudinal studies). In conclusion, it becomes urgent to change the way we do research: the social utility and credibility of intervention science depend on it.