This article defines prescriptive constraints as rules enabling inference engines to check that certain formal objects are used – not just inferred – or not, in certain conditions. It shows why these constraints require not exploiting inheritance mechanisms (or other mechanisms automatically adding relations to objects) during the tests of rule conclusions. It gives a general method to do this and then SPARQL commands to implement this method when the rules are represented via a subclassOf relation from the condition to the conclusion, i.e. as in OWL2-ER. The article illustrates these commands to check some ontology design patterns. More generally, the approach can be used to check the completeness of an ontology, or to represent constraints in an ontology (rather than via requests or ad hoc procedures) to calculate a degree of ontology completeness. The approach can thus help elicitation, modeling or validation of knowledge.