This study empirically assessed the dimensional structure transformation of the top-tier global cities’ city brand influence over six years before and after the COVID-19 outbreak. Based on the newly constructed assessment framework of similarity network analysis, investigations following two paths (i.e., both time-series and cross-sectional similarity network analysis) and six analytical perspectives (i.e., degree of stability, staging, shift, difference, clustering, and representativeness) were performed on the dataset of annual dimensional structures, which was obtained based on machine-learning-driven semantic embedding mining on the full-sample big data of 49,368,287 related global news articles. The study discovered that the transformation stages and direction of the dimensional structures over six years align with the progression and characteristics of the COVID-19 pandemic. Following the COVID-19 outbreak, there was a significant rebound in the differentiation among the dimensional structures and an increased instability of their clusters. The classic tension between brand consistency and relevance also existed in the dimensional structure transformation process. The research findings are valuable for targeted global city branding practices. The newly constructed assessment framework and quantitative results can provide methodological reference and data support for future research in the field.