The foundation of the site's usefulness is that the ratings have some signal. The value of job listings to a Glassdoor user is that they're from companies that Glassdoor rates as positive: an undifferentiated mass of ratings would be a completely useless site. The idea that they have no incentive to improve discriminativity (at the expense of company rating) proves way too much.
Eg by that logic, Yelp would have no incentive to allow bad reviews. Hell, they could just be a listing site that puts "5 stars (10,000 reviews)" next to every listing.
This is true but there is enormous pressure for these businesses to directionally encourage review scores to trend north. Higher reviewed locations and businesses always garner far higher engagement, and this directly impacts top line. As you point out, there is some ephemeral point at which users might begin distrusting the site en masse, but this inflection point is always easy to ignore as "a problem for another day far in the future." In the mean time, tiny but substantial changes occur all the time which nudge that score higher and higher.
In my experience, that macro trust issue is rarely discussed, even though, at some undefined point in the future, it could pose a serious existential threat.
Sure, I agree there's an incentive force pushing scores up. I was responding to a comment asking what possible countervailing force could push scores towards accuracy. The answer is pretty obvious, as the yelp reductio ad absurdum shows.
Eg by that logic, Yelp would have no incentive to allow bad reviews. Hell, they could just be a listing site that puts "5 stars (10,000 reviews)" next to every listing.