System requirements are usually written in natural language, which inevitably is associated with ambiguity, missing details, and inconsistencies. These issues may lead to costly errors if they are detected and fixed in later stages of development. We present an ontology-driven approach for writing verifiable requirements using predefined templates (boilerplates). Our ontologies ensure semantic consistency and support automated detection of ambiguity, incompleteness, and underspecification. The method also supports semi-automated generation of a model with formal semantics and the safety analysis of the system under design. Our approach was applied to real aerospace systems, for which it was demonstrated that it can improve requirements quality, reduce manual effort, and enable early verification.