Legit comment, thanks. What I meant to say is that there's no substitute to empirical data, so you should do the best you can (research and all) but then go back and see what people actually type, which I personally was surprised by.

For example, we had a thought to create app shortcuts from within the search (Facebook also do that I think), but going to the data we saw people didn't really look for features. We could still do it and add education, but the actual data counts a lot here.