Nowadays, the widespread and harmful fake news has generally become a significant consideration. Fake news has the ability to evolve and create huge difficulties for human detection. Therefore, machine learning-based approaches to differentiate fake and real news have emerged as a predominant strategy. Historically, many previous content-based deep learning methods struggle with new types of fake news due to their reliance on static learned patterns. As a result, recent approaches have shifted to using Graph Neural Network (GNN) to include a more comprehensive context and prior knowledge. In this paper, we introduce a novel and robust framework that combines GNN and energy-based learning. Specifically, by leveraging the energy score in updating the trainable cost function, our method maximizes the energy gap between real and fake news, significantly enhancing detection performance. The experimental results demonstrate our promising output compared to that of state-of-the-art methods.