Right, normally polls follow a pretty basic pattern. First you, through a mixture of phonecall and online means, obtain a sample of voters from a given area. You ask some basic identifying questions "Are you registered to vote, have you voted, are you likely to vote" etc. From there, you ask a series of increasingly more specific questions.
This is where the first bit of fuckery begins. The questions are generally -very- misleading, For example: "Do you approve of increased measures to ensure vaccination" (That is a verbatim example from CNN). Frankly, the fact that question -only- got a raw 64% yes response is amazing. It's so god damn broad and open that everyone is like "yes". But it's also priming the pump, we already agreed SOME increased measures are fine. So the next questions are very, very slowly more severe, and lead to additional positive answers.
Now comes the second fuckery. The next bit is to compile the data to see what kind of range of answers you got. The traditional value is to put them into Republican/Democrat/Independent groups. This is where oversampling comes in. Now, ove sampling is -not inherently bad-. What is normally done is you weigh your numbers based on national averages or regional averages, so they balance it out. To massively oversimplify, if you know there is a 50/50 split of Dems to Republicans, but you got 12 Democrats and 8 Republicans you'd weight the democrat responses down to 0.83 and weigh the republican responses up by 1.25 so they both are 'equal' to 10.
The democrats, and CNN, technically do this. But they -massively- overweight Democrats.
This leads to our final bit of fuckery, Time period. CNN in particular is now taking 'polls' over the span of months and compiling them into one to provide a more 'accurate' view. This is bullshit. People's views change over time, it's why we use graphs to indicate changes over time in the day to day. No poll is accurate out to more than a week which is why they are generally shorter. Polls are an indicator of the movement of opinion, not the opinion in and of itself. In isolation, any given poll can only tell you a snapshot -at the time- of an opinion. Having them spread that out over months is going to garble it with utterly useless info and be utterly meaningless.