Let's take another look at the Trump v. Hillary polls.
It looked to me like the polls that nudge people toward choosing one candidate or another--as opposed to giving some other answer like undecided, don't know, etc--tend to yield a smaller gap.
I calculated a correlation between the size of the gap and the percentage of people NOT saying Trump or Hillary for the 13 recent surveys listed on either RealClearPolitics or the Huffington Post. I threw them all in there, including the ones I considered to be not-so-great surveys.
The correlation is .38 which means of moderate magnitude. In other words, when techniques are used to get people to take a side, Hillary's lead decreases. When nudged, the uncommitted tend to go for Trump. To illustrate, 23% of Reuters participants fail to give either candidate as their answer, and this poll shows Trump behind by 7 points. With the UPI/CVOTER poll, which managed to get all but 4% to make a choice, Trump came out on top by one point.
So this seems like evidence that the undecideds are leaning Trump, and in the end, most will vote for him.