Engineering and production in aviation inherently have a disproportionally high documentation requirement – both regarding quantity and quality. Technical writers need software support for the integration of (references to) preexisting textual documents and structured data, e.g. CAD drawings and FEM data from engineering. Thus, these recommendations disburden recurring repetitive documentation tasks. At the heart of such a recommender system, a notion of semantic similarity needs to be established, which is based on balancing the right quantity and right quality of ontological modelling and reasoning, and which covers the linguistic as well as the engineering modelling worlds. It is crucial for the acceptance and daily usage of such a system to mirror the user’s reasoning regarding the quality of the recommendations on different levels: on the syntactic level by NLP standard methods, on the semantic level by ontological reasoning and in-between by allowing users to apply additional filtering on the recommendations, e.g. based on meta-data. We evaluate a prototypical implementation of such a recommender system in a selection of practical use cases in engineering and production of aircraft components and provide lessons learnt for establishing such a system as assistant for everyday, company-wide documentation tasks.