Back in 2010, I did a post on I Write Like which reported the author most similar to a writing sample. I gave it several samples for which it gave me several different authors. The trend I noticed was the topic of the sample seemed to predict the result.
uses linguistic analytics to extract a spectrum of cognitive and social characteristics from the text data that a person generates through blogs, tweets, forum posts, and more.
It matches keywords in the writing to the Big 5 personality test and gives a summary based on it. So, it should be easy to skew too. I was able to find pieces of text from my blog that skewed the scores for all five measures. So, just like the other one, which samples I give it determines my “personality.” Something like FiveLabs’ Facebook Analyzer where it is looking at all or at least a huge sample of my writing probably would work better.