Methods, computer-readable media, and systems for extracting referential keys from a document are provided. A document is parsed to identify at least one key, the key being identified from at least one contextual indication. The key is classified according to a key type, the key type being identified from the contextual indication. The key is extracted and then stored in a location in a structured shell with the location corresponding to the key type. As a result, the key can be found by a search seeking one of the key and the key type allowing a searcher to identify the document from which the key was extracted.