I’m not sure I understand exactly what you’re asking. The sentence you quoted is someone postulating that traffic timing, size, and interactivity analysis can determine both the end site you are using, and infer the content. The person cites some research about the determining the web site you are on. Very popular web sites with known headers and protocols can probably be discerned. For example facebook and instagram are possibilityies, though they litter the connection with ads and content that make the picture noisier. Web apps like gmail and roundcube might be able to be picked out this way too. Or maybe even a banking app. But certainly only in very broad strokes. An adversary certainly isn’t going to be able to pick out your account number, for example. Also, tracing that activity back to you proper will be hugely challenging.