It doesn't use embedded metadata in the file, which is usually pretty low quality. It looks for various identifiers (DOI, ISBN) in the first few pages that can

It doesn't use embedded metadata in the file, which is usually pretty low quality. It looks for various identifiers (DOI, ISBN) in the first few pages that can be used to retrieve high-quality metadata from external services. It also does some analysis to try to identify the title, authors, abstract, etc., and compares those against known metadata for further lookups and/or to supplement the retrieved metadata. (This is a web service [1] because of the database requirements, but we don't log any data about the contents or results of searches, and it's an optional feature.) For extracting metadata from a formatted bibliography you can use AnyStyle [2], which is a separate service written by a Zotero developer. [1] https://github.com/zotero/recognizer-server https://github.com/zotero/recognizer-server [2] https://anystyle.io https://anystyle.io