This research aims at discovering deviations of speech mechanisms that cause errors and mistakes in blogs and online media publications. The deviations of the speech mechanisms are revealed in lexical mischoices, grammatical and discourse errors. We apply semantic and discourse analysis to uncover the errors and mistakes in our dataset (5500 words) of erroneous sentences from Russian blogs and online media. We offer the classification of the mistakes based on the relations between the appropriate and erroneous word forms. We compare the errors and mistakes from our dataset with the deviations from norms in the texts from Corpus of Russian Students Texts. The comparison allows for clarifying the peculiarities of the computer mediated communication norms. We uncover the causes of the mistakes thanks to psycholinguistic analysis of speech production. The bloggers are susceptible to contextual priming. They made wrong lexical choices under the influence of erroneous anticipation of the following element in word combinations. Their underdeveloped skills of verbalizing complex content bring in wordiness, distortion of propositions and logical connections. In the computer-mediated communication, syntactical errors regularly occurs in blogs because of the trend to choose overcomplicated structures. The mistakes in the blogs are mainly caused by hypercorrection and diminished inhibitory control.