Abstract: To clarify the formation and evolution of non-metallic inclusions during fusion welding and their role in microstructural control, this study employs a literature review and mechanism-based synthesis to systematically analyze the sources, types, formation and evolution, microstructural effects, and service-related impacts of inclusions in weld metal. The results show that inclusions mainly originate from pre-existing particles in base and welding materials, deoxidation/desulfurization/nitridation reactions in the weld pool, slag–metal reactions, shielding atmosphere, and external contamination. Their size, composition, morphology, number density, spatial distribution, and interfacial state are jointly controlled by O, S, and N activities, microalloying elements, weld-pool flow, rapid solidification, multipass thermal cycling, and welding parameters. Fine, dispersed, and compositionally suitable composite inclusions can induce intragranular ferrite or acicular ferrite nucleation through lattice mismatch, Mn-depleted zones, interfacial energy reduction, and local strain caused by thermal expansion mismatch, thereby refining weld microstructure. In contrast, coarse, sharp, aggregated, or weakly bonded inclusions tend to trigger cracking, fatigue, hydrogen-induced cracking, and localized corrosion, especially in high-strength steel welds. Therefore, weld-metal inclusion engineering should shift from simply controlling inclusion content to an integrated composition–process–inclusion–microstructure–property strategy oriented toward service performance.