The rapid seismic performance assessment of large structural inventories has become a necessity. This is essential for regional seismic risk assessment and management. However, traditional approaches for estimating the seismic response of structures are either computationally demanding or insufficiently accurate, highlighting the need for new techniques such as data-driven approaches. This paper thus aims to provide a systematic review of machine learning applications in seismic performance assess-ment. By synthesizing trends, key challenges, and future opportunities. A systematic review of 150 peer-reviewed articles published in scopus indexed journals between 2016 and 2025 was conducted using a targeted search string, followed by bibliometric analysis to highlight the research landscape. A classification of ML techniques is pre-sented, followed by a brief overview of the commonly used ML models in seismic per-formance assessment applications. Overall, the analysis of the reviewed papers re-vealed three primary applications: (i) failure mode identification and capacity predic-tion, (ii) seismic demand and damage state prediction, and (iii) seismic response time series prediction. While the findings underscore the potential of ML in advancing seismic performance assessment, several challenges persist, including data scarcity, the black-box nature of ML models, limited generalization capabilities, and high com-putational costs. Potential pathways forward include integrating physics into ML model training, expanding annotated datasets, adapting state of-the-art algorithms for structural engineering applications.