Given a winner's bias in recall, this usually shifts polls towards the loser- which in this case would be Trump.
As such, there is a stark difference in polls that weight on recall versus those that do not.
This is another reason to not expect the same polling error and bias as 2016 and 2020. This is a large change in polling methodology. Two thirds of polls are weighting on recall.
Historically weighting on recall would produce worse results. As Cohn explains, pollsters are justifying the change because they want to make sure they are getting enough Trump voters to not repeat the polling errors of previous elections. But they may be getting too many Trump voters.