Objective: To construct a prognostic model for ovarian cancer based on endoplasmic reticulum stress-related genes (ERSRGs) and systematically evaluate its ability to predict patient survival outcomes and response to immunotherapy. Methods: Transcriptomic data and clinical information of ovarian cancer patients were integrated, and machine learning algorithms were applied to identify ERSRGs associated with prognosis. A risk score model was developed based on the selected key genes, and its predictive performance and robustness were assessed in independent training, validation, and overall cohorts. Single-cell transcriptomic data were further utilized to explore the expression patterns and potential regulatory mechanisms of these key genes within immune cells. Results: Nine key ERSRGs (ERBB2, NHLRC1, CREB3L4, CALR3, MAPK13, OSBPL3, HSD11B2, SLC4A11, GJB1) were identified (P < 0.05). The constructed model effectively stratified patients into high- and low-risk groups, demonstrating a 5-year AUC of 0.663 in the overall cohort. The high-risk group exhibited characteristics of T cell dysfunction and immune escape (P < 0.05). Single-cell RNA sequencing analysis revealed specific expression of these genes in certain immune cell subtypes and tumor cells, suggesting their potential role in modulating the immune microenvironment. Conclusion: A novel prognostic model based on nine ERSRGs was successfully constructed and validated for ovarian cancer, demonstrating robust performance in predicting both prognosis and immunotherapy response. This model provides a theoretical foundation and a potential clinical tool for individualized treatment strategies, showing promising translational value.