The swift expansion of communication channels and social media significantly contributed to the widespread phenomenon of fake news. The detection of fake news has emerged as a critical research area, attracting substantial attention. However, it faces challenges due to limited resources, including datasets and advanced processing or analysis techniques. In this study, we propose a framework to expose fake news that leverages artificial intelligence-based methods. This novel attention weighted logistic algorithm’s distinguishing features include contextual embedding integration, credibility-aware weighting, and the word-support propagation technique. It bridges content analysis with metadata and semantic clustering, making it uniquely robust for fake news detection. This modified algorithm is subsequently compared with existing logistic regression, random force, and naïve Bayes algorithms. The experimental results demonstrate the system’s effectiveness in detecting fake news. The accuracy of naïve Bayes is 81.32%, whereas that of random forest is 84.07%, and that of logistic regression is 87.90%. Therefore, we modified logistic regression and achieved an accuracy of 94.04%.