It's actually more
linguistically correct to use actual
usage as the measure of correctness in language related issues, so this idea makes perfect sense to scientists.
I also use it to decide grammatical issues, but since they usually involve multiword constructs it's generally much harder to find adequate queries and once you have posed a query, to sift out the relevant results. But I highly recommend it - it often clears up some widespread misconceptions on how language is used in the wild.