This idea is from @kostajh. We're going to be offering various algorithmically-driven structured tasks, and the algorithms are all going to perform at different levels of accuracy. Users won't have any idea how accurate they should expect an algorithm to be, and we don't want them accepting too many or too few suggestions because their expectation is way off.
We could tell them in onboarding that they should expect to accept, say, "3 out of 4" suggestions. Ideally, this number should be driven programmatically, as it could vary widely from wiki to wiki.