This article is the 5%.

A more accurate way of detecting "fake news" would be interesting, but I fail to see how such a thing could be designed, past simple detection of wishy-washy and avoidant word patterns.