This came up a few years ago in another forum I've been a part of, on the general issue of recognising anonomous text not for a particular issue in the forum, after something like that happened in the news.

I found I was able to write in a manner which neither person nor software was able to correctly match up with my reference material. About 20% of the people who tried had the same results. Others were matched and were often surprised by what tripped them up when we posted our guesses.

Some people used the very tools under discussion to pre-check their work before posting the anonomous sample. Naturally, they showed non-match in the computer's guess. I furthermore used writing constructs that are among my pet peeves, and a simpler vocabulary (as measured by a reading-level tool), and tripped up the human guessers as well. I think keeping the reading "level" down helped the automatic scans too, since the simpler text has more in common with all text.

BTW, most everyone who tried were successful (published, that is) writers.

—John