Google researchers have introduced Mobility-Embedded POIs (ME-POIs), a new framework that significantly enhances how language models understand physical locations by incorporating real-world movement patterns. The framework combines static text metadata—such as business names and categories—with aggregated and anonymized mobility data like arrival times and stay durations to create a more dynamic understanding of places, moving beyond traditional approaches that rely primarily on static descriptions. The framework delivered substantial performance improvements across multiple metrics. In testing with public benchmark datasets, ME-POIs achieved an 81.9% relative improvement in predicting visit intent, a 75.1% improvement in price level classification, and a 24.7% increase in busyness estimation accuracy. The researchers developed a three-step pipeline—visit alignment, spatial multiscale visit propagation, and text-mobility synergy—that transforms raw geographic and temporal data into numerical vector representations encoding both a place's identity and dynamic functionality. The framework's ability to handle data-sparse locations represents a particularly valuable innovation. While major landmarks and popular businesses generate abundant visit data, small neighborhood shops and newly opened establishments historically suffered from sparse information. ME-POIs addresses this 'long tail' problem by analyzing regional patterns and statistically transferring insights from data-rich neighboring locations, enabling accurate inferences about small businesses even without historical visit data.