google was originally based on pagerank, which was based on the idea that if you analyze the link structure of the web, you can assign quality scores to pages based on number of inbound links, and then use that quality score to propagate a high quality score to other pages that are linked to by pages with high quality scores. in other words: find the pages with reputations you trust, use their opinions to boost the reputations of other pages in the graph.

you could do the same for people. first off, a user looking at a search results page isn't uninformed, there's lots of signal in the results page for a search query: domain name, familiarity/recognition of domain name, abstract text quality (grammar/spelling), abstract text, spamminess, etc. for the trained eye, that's a good amount of signal, but who has a trained eye?

you could, say, have some ground truth rated webpages that you have human raters rate in house, and then you could use this to score actual users on the website in terms of who frequently picks the known best result. now you have a cohort of users who you trust in terms of clicking on quality search results.

now you just pay attention to what this cohort pays attention to and let their clicks materially boost the ranking of results.

this is just one over simplified way, i'm sure they do tons of stuff like this (with tons of other stuff to avoid abuse/seo/etc).