The need for large, high-quality annotated datasets has become critical in the rapidly developing field of artificial intelligence (AI). Manual labeling of data is a major component of traditional supervised learning methods, which are labor-intensive and prone to human error. Automated data annotation attempts to overcome these issues, but current methods frequently fall short in terms of accuracy and consistency. This paper investigates the incorporation of self-supervised learning (SSL) into automated data annotation processes to improve the robustness and reliability of AI models. Without the need for human intervention, SSL generates pseudo-labels by utilizing the inherent structure of data. Our proposed methodology displays considerable increases in model performance and generalization when applied to varied datasets. Experimental results reveal that SSL-based annotation not only decreases labeling costs but also boosts the robustness of AI models against noisy and missing input. This research has broad implications for various AI applications, such as natural language processing and computer vision, among others.