With the rapid rise of internet-based media sharing, the misuse of deepfake videos—highly realistic but manipulated media—has become a critical threat, especially in political, social, and security contexts. Deepfakes are increasingly accessible due to advances in AI and inexpensive computational resources, making their detection a pressing challenge. This survey presents a comprehensive review of current deepfake detection techniques, with a focus on AI-based approaches. We analyse state-of-the-art methods, particularly those employing Convolutional Neural Networks (CNNs) like ResNet50, Recurrent Neural Networks (RNNs), and hybrid architectures. Additionally, we examine various classifiers such as Random Forest, Support Vector Machine, and Logistic Regression used in detection pipelines. Through comparative analysis of techniques, datasets, and performance metrics, this paper highlights current trends, key challenges, and potential directions for future research in deepfake video detection.