<p>Accurate prediction of structural seismic response is essential for earthquake-resistant design, but conventional finite-element (FE) time-history analyses are computationally intensive, limiting large-scale or real-time use. In recent years, machine learning (ML) has emerged as a promising alternative, capable of learning complex nonlinear earthquake-structure relationships and providing rapid inferences. This review addresses a critical gap by combining a scientometric analysis with a systematic evaluation of the ML-SSRP (Machine Learning for Structural Seismic Response Prediction) literature. The scientometric study (Scopus, 2019–2025) maps research trends, collaboration patterns, and thematic developments, revealing rapid growth in the field, dominated by contributions from China and the USA, and a strong emphasis on data-driven prediction of reinforced-concrete systems. In parallel, 65 representative studies are reviewed to synthesize methodological developments across diverse ML architectures (e.g., ANNs, CNNs, LSTMs, transformers, ensembles) and structural typologies. Input features, model structures, seismic contexts, and performance metrics are systematically analyzed. The findings demonstrate that ML-based models consistently deliver high predictive accuracy with significantly reduced computational time compared to traditional simulations. However, challenges such as dependence on idealized training datasets, limited generalizability, and lack of interpretability persist. Promising directions include hybrid and physics-informed frameworks that aim to enhance robustness and adaptability. By integrating bibliometric mapping with in-depth technical analysis, this review provides a comprehensive overview of the current landscape, identifies key strengths and limitations of existing approaches, and outlines strategic priorities to advance ML-driven seismic response prediction in structural and earthquake engineering.</p>