This article presents a systematic literature review aimed at analyzing how data from social media platforms — particularly X (formerly Twitter)— have been employed in the formulation of predictive models applied to health surveillance, using machine learning techniques. Twenty-three studies published between 2018 and 2021 were examined, covering the period immediately before and after the COVID-19 pandemic. These studies applied analytical algorithms to georeferenced posts in order to detect outbreaks, monitor epidemics, and support decision-making processes in public health. The findings suggest that such models may anticipate risk clusters up to four weeks in advance, enabling more timely responses from health authorities. However, significant limitations to the adoption of these systems were identified, such as the scarcity of geolocated data, susceptibility to informational noise, and technical and ethical constraints related to automated natural language processing. The review indicates that, although the applicability of these solutions is expanding, their institutional consolidation depends on integrated strategies for validation, algorithmic governance, and the enhancement of digital infrastructure. This study seeks to contribute to the field of Digital Health by offering a critical perspective on the pathways and barriers associated with the use of social media and artificial intelligence in health surveillance.