The proliferation of deepfake technologies has created serious problems for the authenticity of social media material, particularly textual content. Applying deep learning and rapid text embeddings, this research demonstrates a way to identify tweets created by machines. We build a dataset that includes real and fake tweets, showcasing various linguistic styles. Quick text embeddings allow a deep learning model to grasp semantic subtleties, which in turn allows for efficient feature extraction. By using this dataset for training and validation, the model is able to reliably and accurately differentiate between real and fake material. The results show that this method helps in the fight against social media disinformation by making deepfake text easier to spot. To protect online conversation against deepfake dangers, automated tools are crucial, according to this research.