I looked at a ton of polls in 2016. The problem wasn’t only the media’s “projections,” it was also pollsters’ sampling errors, turnout models, and various formulas/models they used to “normalize” poll results. If you get into the raw numbers and data for each poll, you can see the errors up close. I recall one poll that had about 70% Democratic respondents to 20% Republican, with rest undecided/independent. More than half of the responses were via landline. The pollsters attempted to extrapolate a representative sample from its original sample using some complicated formula based on turnout, party registration, and the state’s voting history. What they should have done is find a way to poll the representative sample in the first place. Anyway, the poll showed Hilllary nearly beating Trump in some red state (TX, AZ perhaps). We all know how that turned out.
Anyway, as relevant to the OP, when in 2016 did Dems finally realize they were in trouble in MI, WI, and MN? Do you remember? It was in November, like a week before the vote. Prior to that, pollsters had all but ignored those states. My point is this: not polling in inconvenient areas (notwithstanding polling errors) is the Dems’ strategy for maintaining enthusiasm. It’s also the recipe for a “November surprise.”