<p>Fake news is a significant challenge in today’s digital age. Its rapid spread caused by the massive adoption of social media, poses significant challenges to the credibility of information and news impacting the stability of societies. Therefore comes the necessity for effective fake news detection methods. Our paper introduces UFNAC (Unsupervised Fake News lAbeling and Classification), a framework that addresses the limitations of supervised models’ dependency on labeled datasets for fake news detection. Our solution uses unsupervised ensemble learning techniques to label unlabeled datasets, which are then used to train a supervised Convolutional Neural Networks (CNN) classifier using DistilBERT-generated embeddings for feature extraction. By integrating unsupervised and supervised methods, UFNAC expands the applicability of fake news detection systems to scenarios with limited labeled data. The contributions of this research include exploring unsupervised labeling techniques to annotate unlabeled datasets. This strategy expands the options for utilizing real-world data, which is often unstructured and unlabeled, enabling more practical and scalable applications. These advancements lay the groundwork for creating reliable tools to combat misinformation effectively across diverse contexts.</p>