First, start with a whitelist. Hand pick high quality publications, and rank them towards the top. This may tilt results back towards institutions, and away from blogs.
Second, punish similarity. If everybody is reposting AP or Reuters without any additional information, consider them a dupe and don't list them. They can run their portals, but they don't need to show up in search.
It's come up multiple times in this thread, car manuals is a good example. They would be better off throwing away every result they have and hand indexing the good information, than what gets returned right now.
Recipes in particular have turned into a giant story about the way grandma used to do it with a picture followed by the same couple variants with different proportions. Pick winners by hand.
Someone has a finance question, just put boggleheads at the top, instead of whichever 59 affiliate credit card sites sprung up.
Need health advice? Put examine.com at the top above WebMD and healthline. Why? Because a human exper compared them and decided examine is a better first result. You could comb through tens of thousands of sites with a team of hundreds of people, something Google easily has at its disposal. What PageRank had, that seems to be missing now, is a seed of "we trust these most" and let the network grow from there. It tried to find expertise, instead of clickability. It was about getting you the best information first.