Abstract <p>We have analyzed over 80 scientific articles from around the world on the topic of using convolutional neural networks in seismology. This review is structured as analysis of the distribution of these publications across several groups of characteristics. We have previously formulated our vision of the scientific field under consideration, focusing on such key points as the problem of the volume of data sufficient for high-quality training of neural networks; difficulties with formalizing the choice of a suitable architecture for the created neural networks; and lack of consensus among researchers on optimal methods of preliminary data preparation. Before the actual analysis of publications, the methodology of the search process is described and list of scientific databases containing links to publications from around the world is provided. The review begins with study of the long-term dynamics of publication activity on this topic, identifying specific stages when scientific progress proceeded at different speeds. Interpretation of the identified patterns can be useful in predicting the further development of this direction. Next, the distribution of the collected publications is analyzed and interpreted according to the tasks that seismologists set for the convolutional neural networks they create. As a result, the most relevant topics of our time and those that have not yet received the attention of researchers are identified. The review concludes with analysis of the problem of data preparation before loading it into the created neural networks for training and work. Different methods of data pre-preparation are analyzed by comparing their advantages and disadvantages.</p>