Clients occasionally suggest to me that I start early on an indexing project, before the page proofs are finalized, by making a list of the terms that will appear in the index. The reasoning behind this is that when the page proofs are ready, I can search the PDF for these terms and quickly and easily fill in the page numbers. This is similar to when I hear people say that ebooks don’t need indexes, because the search function is sufficient.
This approach works for a limited number of books, such as certain reference or other fact-based books where the goal of the index is to point the reader to the facts. Name and title indexes could also be constructed using this method. Beyond this narrow range of projects, however, I think that the word list approach to indexing misses the point of what a good index seeks to accomplish.
Yes, searchable terms, if relevant to the reader, should be indexed, but a book is often so much more than its facts or names. A good biography or history will have a narrative. A good scholarly work will have an argument. A good cookbook index will have categories such as breakfast, lunch, and dessert, or spicy, savoury, and sweet, that may not be immediately obvious in the text but still relevant and helpful to the reader. These are the types of entries that cannot easily be found using the search function. If the index is truly a map, a reflection of the text, these entries need to be present.
I also question the assumption that starting with a word list gets the job done faster. In my work, I try to read the text and create the index entries at the same time. I am reading more slowly for comprehension as opposed to just reading for the key words, but overall I find that this is faster than reading once to make a word list or to mark up proofs, and then a second time (even if just using the search function), to input the entries. Of course, some books are more difficult to understand and require re-reading sections two or three times, and when I edit the index I do go back and check entries for accuracy–the one pass approach doesn’t always work as planned–but in my experience it is more efficient. Add in subheadings and the thematic and conceptual dimension of the index, being able to see the context while reading is far more helpful than using a search function.