When people bring up the fact that the Democratic candidates are all way above Trump in the polls, a common response is that Hillary was ahead in the polls but she lost but she lost so they don't mean anything. But the polls give a probability of a candidate winning, not a certainty and yes some sources were claiming that Hillary had a 95% chance of winning but that has to be seen in context of how many elections the polls predicted. The human mind is notoriously bad at is statistics; most notably how fast the odds of an improbable event happening build up as the number of replicates increases. For example, if a coin is loaded to come up heads 99.9% of the time and you flip that coin 10 times, the odds of at least one of those flips coming up tails is approximately 1 in 2. So, even if the polls were 99% accurate at determining who would win the election, that they'd fail to do so on some elections is to be expected. Now due to the electoral college system that the US has, they are less accurate than they are elsewhere (and can be improved by polling specific battleground districts [https://www.dataforprogress.org/memos/battleground-favorability]) but that doesn't mean that they're completely useless.