The rapid spread of misinformation on social media platforms, especially Twitter, presents a challenge in the digital age. Traditional fact-checking struggles with the volume and speed of misinformation, while existing detection systems often focus solely on linguistic features, ignoring factors like source credibility, user interactions, and context. Current automated systems also lack the accuracy to differentiate between genuine and fake news, resulting in high rates of false positives and negatives. This study investigates the creation of a Twitter bot for detecting fake news using deep learning methodologies. The research assessed the performance of BERT, CNN, and Bi-LSTM models, along with an ensemble model combining their strengths. The TruthSeeker dataset was used for training and testing. The ensemble model leverages BERT's contextual understanding, CNN’s feature extraction, and Bi-LSTM’s sequence learning to improve detection accuracy. The Twitter bot integrates this ensemble model via the Twitter API for real-time detection of fake news. Results show the ensemble model significantly outperformed individual models and existing systems, achieving an accuracy of 98.24%, recall of 98.14%, precision of 98.42%, and an F1-score of 98.24%. These findings highlight that combining multiple models can offer an effective solution for real-time detection of misinformation, contributing to efforts to combat fake news on social media platforms.