Abstract: Geospatial data constitute a fundamental infrastructure for digital Earth, smart cities, and national defense, and their currency directly determines the reliability and application value of the decisions they support. Driven by advances in artificial intelligence, Earth observation, and map generalization, geospatial data updating has made notable progress in change perception, multi-source fusion, and cross-scale propagation,while still facing considerable challenges. Following the main thread of change-information discovery—multi-source data fusion—multi-scale updating, we review recent developments and outlines future directions. First, change-information discovery is examined from two complementary perspectives: Remote-sensing image change detection—covering direct comparison, post-classification comparison, and large-model-driven detection—and vector data matching, categorized into same-entity matching, same-type (group/pattern) matching, and cross-type joint matching. Second, the key techniques of multi-source data fusion and consistency processing are organized along four dimensions: spatio-temporal reference, geometry, semantics, and spatial relations (topological and directional). Third, taking map-generalization operators (selection, simplification, aggregation, typification, etc) as the core, we analyze the fundamental methods of scale transformation and three typical updating modes: master-scale derivation, multi-scale collaboration, and incremental cascading updating—together with graphic-difference-based, geo-event-based, and artificial intelligence (AI)-model-based incremental updating methods. Finally, in light of emerging directions such as large AI models, multi-source collaborative updating, and continuous-scale representation, it discusses the trends and challenges of geospatial data updating toward intelligence, real-time performance, and service orientation. We further argue that a promising breakthrough may lie in embedding multi-scale association and update-oriented metadata at the data-production stage, rather than relying solely on post hoc detection, matching, and fusion.