I guess you could skirt around this by using something to tag the various parts of speech in your original text (using something like Python's NLTK) and replace them with randomly picked synonyms from a thesaurus?
Pretty sure it would obscure the original writer although possibly at the cost of obscuring the original meaning.
I think what we’re concluding here is that using Google to obscure the linguistic style is flawed, because a state actor could obtain the original linguistic style from Google records, or from their own records of snooped traffic.
In other words: the blog should find a way to obscure linguistic style offline.
Wait, why ask Google, they probably can just look in their own surveillance database.
Geez, if Google Translate queries are logged, that's... a lot of information.