On top of what has already been mentioned, there is legitimate criticism about what Nate decides to include in his model. They will include a PPP poll without question, unless that poll is paid for by a specific campaign, candidate, or organization. That's a completely understandable policy, even if I disagree with it. At the same time, they will happily include polls that are by questionable pollsters, such as Remington. (You could pull better numbers out of a hat). They include the LA Times Poll, but have to "adjust it" by nearly 5 points each day. At that point, why? Why bother?
They include these ultra small polls that aren't actually state polls at all, but are subets of a national tracking poll divided by state. Some of the polls had a sample size of like 100, but were weighted more heavily than other polls. Nate went on a Twitter rant about it, saying that they should be included, only to go back on that a few days later when he was corrected on them.
I understand his idea that you use every data point and just hope it works out in the end. The problem is, this year there aren't a lot of high quality polls. His model is flooded with shit level polling that is probably gunking it all up.
Also, he is a terrible pundit. He is so thin skinned, and his writing is terrible. Harry Enten is the best part of 538. He's able to walk the line between snarky and bitchy. Nate is not. A few months ago, someone asked a question as to why Nate's model was doing a specific thing, and he blew up at them on Twitter.
There are legit criticisms of Nate. Now, saying he is doing what he does to his model for clicks is wrong. Saying he writes shitty articles to get people to click on his website isn't inaccurate.