The rising level of fake profile creation and coordinated bot-driven activities raises a level of demand for effective detection systems in social media sites. The old approaches are extremely rare because they usually depend on individual profiles or temporal patterns, which can hardly capture the complexity of the social structure in which these accounts really operate. It might miss coordinated behavior and is unable to recognize fake profiles embedding themselves in social networks of legitimate users. We propose a graph convolutional network, GCN, for social network analysis, tailored to detect fake profiles and bot communities which exploit the innate structure of the graph representing social platforms. Social networks are inherently graph-structured: Nodes are users, and edges follow interactions-whatever follows, likes, or retweets are the edges. GCNs can effectively represent these ties, and are able to capture critical local and global connectivity patterns important in detecting abnormal behavior. In the proposed approach, we rely on the features of user profile such as follower count and engagement rate as node attributes and interaction data as edges to classify profiles as original, fake, or bot-driven in process. The community detection also comes in handy with the GCN framework that identifies clusters of coordinated fake profiles operating within the network. It introduces layered analysis that takes into account more than the restrictions of the traditional models and uses individual features coupled with structural connectivity for better detection performance. The obtained numerical result strongly supports the validity of the presented approach, since the high accuracy (96–98%), precision (0.92–0.94), and recall values (0.91–0.93) with an F1 score of 0.92–0.93 were obtained. The work significantly improves the detection precision and recall for bots, thus providing a good tool to significantly limit the spread of disinformation and ensure the integrity of social networks.