And census tools like Geocorr can translate between zip/zcta and other geographies, weighted by population or land area, without doing any spatial calculations.
Don't most search engines use an inverted index to find the similarity between the query vector and the document vectors? (instead of doing the dot product with every document)