I'll be the first to admit that this requires some knowledge of CSS. Or, in my case, turned out to be how I learned most of what I know about it ;-)
There are a few general observations I've made over the years:
1. A whole lot of Web design is absolute crap. Or as I put it, "Web design isn't the solution, Web design is the problem."
2. Less is virtually always more. I'm constantly shocked at how much calmer the experience of reading a site to which sidebars, recommendations, social-media link-litter, advertising, and of course, autoplay video and audio, have been removed.
3. There's a massive amount of shitty HTML out there.
4. Content extraction ... can work, and surprisingly well for some sites. I'd built an html-xml parser for the Washington Post which reduces news articles to about 1% of their original size, and tremendously improves readability. Those tend to have to be configured for specific sites or content engines, however.
5. That said, there's a lot of re-use of content engines, and some rules can prove quite generalisable. But ...
6. ... There's also a lot of churn, and rules which work at one point in time can fail spectacularly at another. Determining what version of the generator is being used for a specific request is ... difficult.
7. Often the most dramatic improvement is simply to dump plain text of the main meat of an article and re-apply formatting to that. This is of course manual, but adaptive and ... reasonably quick. What also becomes rapidly apparent is that site styling and markup offer very little additional benefit or structure to the core content.
8. For almost all commercial sites, there's virtually no use of bold/strong, and little of emphasis, outside of advertising or ancillary content. Simply nuking and tagged-content is remarkably effective at trimming crud. Often it's the only reliable way to do so.