Typically dynamic content doesn't change from second to second, it changes after 5 minutes or an hour or 1 day, actually it is extremely site specific too. But

Typically dynamic content doesn't change from second to second, it changes after 5 minutes or an hour or 1 day, actually it is extremely site specific too. But I do like your idea. To go a bit further on your idea - you could apply machine learning to analyse the changes. So for example, ML could determine what is probably the "content area" of the page simply by having built out a NN for each website that self-expires the training data at about 1 month (to account for redesigns over time). The major problem will still be "ads" in the middle of the content, especially odd scroll designs ads that have a different "picture" at each scroll position, as well as video ads that are likely to be different at each screen shot. Another form of ad being the "linked" words like when you see random words of the paragraphs becoming links that go to shitty websites that define the word but show a bunch of other ads. I suppose Google could simply install UBlock in it's training data collector harness to help with that stuff. >()