Today, the pervasive presence of misinformation on social media makes it necessary to determine the truthfulness of information. This task involves categorizing information as either true or false. Given the immense volume of data on the Web, this task becomes particularly time-consuming and very difficult for humans. The use of deep learning models is essential to automatically solve this with a considerable degree of efficiency. This study introduces hybrid deep learning algorithms that merge two neural network models. The strategy combines a ConvNet and LSTM, yielding performance that surpasses the techniques reported in prior research. When applied to fake news detection, the proposed models achieve high performance on FA-KES and ISOT datasets, two specific datasets designed for the study and detection of fake news. Additionally, experiments involving the generalization of models yielding results surpassing those of other hybridization models found in the literature.