Many people can scan pages at the rate of a few seconds per page to pull out keywords. They are relying on the same ability to fuzzy-match the structure of a sentence, looking for the topic, beginning of a dependent clause, Personal Names, and so on. I spent far too much time doing that before we could ctrl-F everything. It's not all that reliable but it does seem to suggest people can match against text in massive parallel. And which word percolates first up to consciousness when confronted with a block of text is not random.