Abstract. Humans require automated support to profit from the wealth of data nowadays available on the web. To that end, the linked open data initiative and others have been asking data providers to publish structured, semantically annotated data. Small data providers, such as most UK real-estate agencies, however, are overburdened with this taskoften just starting to move from simple, table-or list-like directories to web applications with rich interfaces.We argue that fully automated extraction of structured data can help resolve this dilemma. Ironically, automated data extraction has seen a recent revival thanks to ontologies and linked open data to guide data extraction. First results from the DIADEM project illustrate that high quality, fully automated data extraction at a web scale is possible, if we combine domain ontologies with a phenomenology describing the representation of domain concepts. We briefly summarise the DIADEM project and discuss a few preliminary results.