Yes, they can....they just can't auto dial them. Most of the larger polling companines do call a small number of cell customers to get a feel for where they may be voting(kind of a poll within a poll) then assign that number to represent the percentage(7.9%+/-) of registered voters who only use cell phones.

The Pew research center has a study on the whole thing but I don't have a link.

Yes, all of the polling companies attempt to adjust for the cell phone issue (only small polls can even try to completely include cell phones since not using auto dialers makes sample size for large polls extremely difficult).

But in the end all they can do is look at past behavior and make educated guesses to extrapolate. The same issue exists in the "likely voter" models they use since generally they discount youth voting to a degree since the under 25 crowd never votes at the rate they say they will.

Pollsters though have to try to take into account, though, that through the primary season young voters deviated strongly from past models. Will that continue or will the actual election regress to the mean.

So there are a lot of issues that could invalidate the polling. The polling companies are aware of them and try to correct for them. Sometimes they do well and sometimes they don't. Sometimes they just get unlucky (good model produces an outlier) or get lucky (bad models can still give what turns out to have been right).