It can be done relatively easy. But the problem is, those opinions, sarcasm, emotions and other fuckery isn't just a visual problem. It affects the actual content of the article too. And since AI doesn't know what's true or false you'll just end up with a clean, easy to read article, that is still fake news.
It's like comparing a radical tabloid to CNN. CNN is clearly more dangerous because they use less sensationalism than the tabloid, less clickbait, less spelling errors, and such which makes them look more trustworthy. Even though it's still very fake news.