The article introduces ensemble methods to enhance prediction accuracy and evaluates various classifiers using separate datasets. The Naive Bayesian, Passive-aggressive, Linear Support Vector, Logistic Regression, k-nearest neighbors, and Random forest classifiers are assessed. Ensembles comprising combinations of these classifiers are also tested. Results indicate high accuracy for individual classifiers, with ensembles further improving performance. Research results present classification outcomes, showcasing the effectiveness of different methods. The research emphasizes the significance of ensemble methods in fake news detection, providing insights for future research. For TF-IDF Vectorization, the Support Vector Machine (SVM) classifier stands out with the highest mean accuracy of 95.74%. This indicates that SVM achieved the highest proportion of correct predictions compared to other classifiers when trained on TF-IDF transformed data. When utilizing Hashing Vectorization, the SVM classifier maintained its superior performance, achieving the highest mean accuracy of 97.26%. Ensemble methods, specifically Voting Ensemble 3 (Ens3-SVM+PA+LR), exhibited noteworthy performance, particularly with Hashing Vectorization. This ensemble achieved a mean accuracy of 96.93%. The main idea of the method is to analyze a purely news text, without extraneous information (dates of publication, website names, additional media). The text is analyzed according to three metrics separately: truthfulness of the news, satire, or hate speech. Datasets of the Kaggle service were used to train the models according to metrics, and arbitrarily selected texts of news and comments were used for testing in “real conditions”. The structure of these datasets is text and a binary label in another column corresponding to the criterion. The dataset included 6,335 lines of news texts and true or false labels. The satire dataset is a combination of two separate datasets, one from the BBC news service and the other from the humor Onion.