Good question. Actually we have a technology that analyses semantically the webpage and determines the meaning of each elements.
Basically it's a bottom-up approach, we first try to guess the meaning of an element; then we take one step back and try to find contextual information that helps us refine the meaning.
We thought about the crowdsourcing approach, but it requires that the few first users have a shitty experience (because Dashlane isn't working as expected on these sites). That's why we prefered a more generic approach.
But it makes sense. Though, how do you deal with false positive ?
We prefer to rely on our semantic backend, but if we implement a crowdsourced method, I think it would be more to spare computing power than for the quality of our results (which are already quite impressive).
Just imagine the case where a site changes drastically, all the crowdsourced results would be out of date, how do you detect that ? Because a user signaled it ? What if it has been incorreclty signaled ? Sure you could set a sort of treshold, but that means that a bunch of users will have a poor user experience in the meantime.