Content verification studies aim to employ methodologies to identify deceptive contents on social media. Some of the proposed methods include the use of source credibility, source rating and detection. Unlike other deceptive instances on online platforms like review manipulations, where the outcome variables are dichotomous, fake news and real news do not exist in isolation. However, studies on content verification have since treated the concept of fake news as a binary classification task. We bridge the gap in literature by analyzing fake versus real news and then introducing a third class into our nomology. We propose an innovative hybrid approach based on a multi-class algorithm that leverages the use of stacked gradient boosting ensemble and random forests to detect false, real and tweets that can be classified as “noise”. We show that although a classifier's efficacy may suffer after the introduction of the third class, its overall utility is improved.