This is pretty important. Polls just reset each time to correct what they missed the previous time. It's hardly ever asked why they were wrong the previous time.
In 2012, Romney was polled as likely to win two or three of the rust belt states. Turned out a load of people they thought would vote for him didn't, Obama overperformed his polls and his -1 turned into +4 when the results happened. So the polls in 2016 (wrongly) assumed that particular demographic in that particular region wouldn't bother again. But they did, and the polls were wrong. They were no more wrong than they were in 2012, but because it didn't impact the overall predicted winner you didn't get smug pundits crowing over polls being terrible and pointless.
This year they'll get something else wrong, and it won't be the same thing that gave the 2016 result. But it is pretty difficult to see what that something else is that means Trump (legitimately) wins.