Fact-checking is hard. It's hard for humans, but it turns out that it's even harder for AI. Tracking sources isn't enough; you have to categorize whether that source is, in itself, reliable. Some of that is about pattern recognition, knowing whether a source is likely to be a good original or not, and that judgement call is really hard for machines.
"While a 27 percent accuracy rate for a deception detector isn't going to singlehandedly put a stop to the proliferation of fake news any time soon, it underscores the difficulty of teaching a machine to recognize deception."