I recently read that polling response rates are now under 1%. There is almost nothing 'random' about who and why some people still answer a phone or text. So they are no longer actually random sampling, what they are doing is regressing the survey data they can get to correlate well with what they expect as a reasonable outcome or past outcomes and then reporting that their poll is reproducing a reasonable outcome. But any idea that they are getting a representative sample of the population - which is the fundamental assumption undergirding sampled polling, is pure fantasy today. What they are doing will work as long as change remains incremental, because linearity will always hold well enough over small increments. But they are guaranteed to miss anything big happening. Which is not to argue that it is, only they the polls are locked into methodology making them likely to miss it.