Maybe it’s like that trick where if a thousand people guess the amount of beans in a jar almost all of them will be wrong but their average will be very close to, if not, correct.

they probably weight documentation higher

They're applying trust factors similar to PageRank.

Possibly, but at a semantic level.