There’s been some talk about this in other threads, but I’m still confused how it could tell the difference between a voting ring and a legitimately popular article without having many false positives.

I’m curious about this too, but I don’t think they’ll talk openly about it. For some hints, you could probably look at what data Google reCaptcha collects. The newest version (v3) doesn’t even require a user to click anything except a single button.

I bet Reddit and HN use timing, browsing time, clicks, and mouse movements, among other things to tell if a lot of visitors are just coming to the site to upvote a single story or are genuinely browsing the site before coming across something interesting.

I avoid talking about because I don't want people getting ideas, but it's easy to detect voting rings with HTTP referers and voting history alone

Content marketing is usually easy enough to tell; even if upvoted by a voting ring, it'll be mercilessly flagged down.

The flip side I don’t hear about is using up/down votes to shape discussions. You’ll often read discussion on the same topic with drastically different sentiment at the top.