A common challenge in survival analysis is the presence of (right) censored instances, where the event of interest does not occur within the study period. The censoring rate varies across application domains and can significantly influence model performance. To extend, develop, and test new models in a specific domain, it is crucial to simulate survival datasets with a desired censoring rate. Existing approaches often rely on arbitrary censoring time generation or simplified parametric models (e.g., exponential distribution), which do not necessarily reflect the censoring characteristics observed in real-world datasets. In this paper, we propose a novel approach for simulating semi-synthetic survival datasets with predefined censoring rates. Our method ensures a close approximation to the original censoring distribution while allowing precise control over the censoring rate. We demonstrate that our approach effectively generates realistic censoring times by leveraging the relationship between censoring and covariates, outperforming existing methods in terms of distributional fidelity. Quantitative evaluations using the Wasserstein distance assess how well the censoring times generated by our method align with a reference distribution that represents the real censoring mechanism. We compare this distance to those obtained with existing approaches, demonstrating that our method produces censoring times that more faithfully reflect real-world censoring patterns.