The number and size of RDF knowledge graphs are constantly increasing. As a result, data processing for agents (automated or human) is becoming more and more difficult. If several tools can be used for a given task, but each depends to varying degrees on the semantics available in the knowledge graph, then it is important to have an overview before the graph to select the best tool for that task. We conducted a large-scale in-depth study to verify the presence of semantics in knowledge graphs currently published in the Linked Data. Although some knowledge graphs use OWL 2 semantics, many do not do so or only partially. We therefore propose an approach that, based on statistics, instantiates an ontology that facilitates the selection of the most suitable tool for a given task based on the use of OWL 2 semantics.