The rapid advancement of computational techniques, particularly Generative Adversarial Networks (GANs), has facilitated to generate the exceptionally convincing fake content with far-reaching societal consequences. Numerous innovative methods for altering facial features in video content have been created and broadly shared. Deepfake algorithms can generate fake videos that are so convincing that distinguishing them from real ones becomes challenging. This phenomenon has facilitated various crimes, including sextortion, digital fraud, and the spread of fake news. Platforms like WhatsApp and Instagram are commonly used to disseminate these fake videos. It has been observed that these platforms compress the data, meaning that when any video or image is shared, it undergoes compression. This study examines the effect of compression on the detection of Deepfake videos, focusing specifically on faces. EfficientNet is utilized to evaluate the accuracy of a DCT-based lossy compression technique with varying degrees of compression. Additionally, the study explores the impact of compression when videos are shared through WhatsApp and Instagram. The experiments were conducted using the publicly available DFDC dataset. The results indicate a significant decline in accuracy due to compression compared to uncompressed videos.