Steven Johnson is a prolific writer who has written about how he uses DEVONthink databases to support his work. He advocates entering text “snippets” extracted or summarized from books or articles, each of which concisely presents a single concept or fact (and includes source information). And he has concluded that such snippets should be no longer than about 500 words. When he’s writing a new book he finds that exploring the “snippets” database constructed for that project is very efficient, including his uses of See Also.
I’m certain that approach works well for Johnson.
But my databases do not consist just of snippets, and never will. For one thing, Johnson uses paid research assistants to pour through libraries and journals and create those snippets to go into his databases. He’s a well-known author and can justify this as a business expense. I don’t have paid assistants, and I don’t have the time to go through the long documents in my databases and distill them into snippets.
Some have interpreted Johnson’s advocacy of snippets by splitting long documents into “chunks” of 4 or 5 hundred words, either manually or using a script to do that. I’m not going to do that, and I’m quite sure that’s not what Johnson expects to get from his paid research assistants, from whom he expects perhaps just one or a few snippets from a book or article. There’s no guarantee that such a mechanical approach would capture a “whole” fact or idea per snippet, and a high likelihood that the conceptual gem contained in the document will be split among two or three snippets, perhaps in a confusing way.
Finally, I’m a nut about not vandalizing my collections of references. I don’t split, highlight or mark up my documents. If I want to mark up a document, I’ll do that on a duplicate and then delete it when I’ve gotten what I want from it.
My approach, when a search or See Also list presents a long document that might be interesting, is to examine it for possible use. If it seems worthwhile, I’ll create a new rich text note (probably the “Annotation” smart template, though I may create others as well) in which I enter notes or brief excerpts, with links to a particular page in a PDF, or a “cue string” that allows me to do a Lookup search that will be highlighted in the search result.
Such notes become rather like Johnson’s snippets, as they condense and point to particular facts or concepts in my references, and I keep many of them in my databases for that reason. My main database contains about 25,000 references and about 5,000 notes. Many of those notes point to multiple references via hyperlinks.
In the past I’ve been a research assistant with duties like Johnson’s assistants, and later had my own research assistants. Nowadays, I maintain that DEVONthink is the best research assistant I’ve ever had, as well as the most inexpensive one. 
The collection of reference materials in my main database has been built up over the years and is dynamic, with new content added (perhaps as the result of DEVONagent searches for a topic, as well as through new articles, reports and books of interest from journals and other sources). From time to time I may weed out items, or even put them into groups labeled “Nutty” or “Junk Science”.
how much time did that take?