One thing to keep in mind with, from what I've seen, nearly all the polls at this point is...
The results being touted are within the margin of error.
Stated differently, the results may vary.
From the article I posted in #1...
... The Bloomberg News/Morning Consult poll surveyed 4,932 registered voters in seven swing states: 796 registered voters in Arizona, 788 in Georgia, 698 in Michigan, 447 in Nevada, 699 in North Carolina, 807 in Pennsylvania and 697 in Wisconsin. The surveys were conducted online beginning March 8, and concluded March 14 in Arizona and Wisconsin, March 15 in Nevada, and March 12 in the remaining states.
The aggregated data across the seven swing states were weighted to approximate a target sample of swing-state registered voters based on gender, age, race/ethnicity, marital status, home ownership, 2020 presidential vote and state. State-level data were weighted to approximate a target sample of registered voters in the respective state based on gender, age, race/ethnicity, marital status, home ownership, and 2020 presidential vote.
The margin of error is plus or minus 1 percentage point across the seven states; 3 percentage points in Arizona, Georgia and Pennsylvania; 4 percentage points in Michigan, North Carolina, and Wisconsin, and 5 percentage points in Nevada. ...
From my experience in market research, that is an excellent description of the poll.
Notice the margins of error in that last paragraph.
Also notice the frame of the poll, "... surveyed 4,932 registered voters in seven swing states ..."
This far out from Election Day, the frame of the polling tends to be "registered voters." As we get closer to Election Day, you'll start seeing the frame as being "likely voters."
So there is that also.
(Apologies for dumping marketing research terminology on y'all....)