This paper introduces a novel PCA sweep procedure for Supervised Semantic Differential (SSD), a method modeling how text meaning varies with individual differences. The sweep systematically selects the optimal number of PCA components to ensure interpretable and stable semantic gradients, illustrated through a case study on AI discourse related to narcissism.