analytics toward continuous, personalised and consumer-facing engagement, yet the consumer-research literature on this shift remains fragmented. This review integrates peer-reviewed clinical, information-systems and consumer-behaviour evidence with recent industry surveys to examine how ML is used along the consumer health journey and how consumers respond to it. The present scenario shows credible progress in image-based diagnosis, wearable-enabled screening, risk stratification, care navigation and conversational support, including large-scale evidence that a smartwatch algorithm can flag atrial fibrillation and that chatbot answers to patient questions can be rated higher in quality and empathy than physician replies in a public forum. It also shows persistent weaknesses: proxy-label bias that under-identifies Black patients for extra care, proprietary models that perform worse in external validation than reported, consumer resistance rooted in "uniqueness neglect", and low willingness to share health data with technology firms. The review proposes a conceptual framework in which a trust layer of consent, fairness, validation and human oversight mediates between prediction and consumer action. Future perspectives address generative AI, federated learning, adaptive interventions, consumer co-design and regulation, and a research agenda is set out for consumer researchers. The review concludes that consumer-centric ML will be judged less by predictive accuracy than by whether consumers understand, control and benefit from it.