There have definitely been periods when polls were quite skewed because they were conducted by phone and they over-represented people with landlines. I guess because it’s harder to get a “phone book” of cell numbers. This skewed actual outcomes of the polls.
I think they still have similar issues. They probably underrepresent people who change phones often or just don’t use the telephone to communicate. Sometimes they’re conducted online and recruited over email. There are all different ways that that can skew more to one type of person.
And when they say they have such and such a confidence interval, that does not mean they are that confident they’ve controlled for all the biases. It’s just simple statistics math about sample sizes.
I think we have seen very clearly in recent presidential elections how unreliable polls are. Not only do people answer them differently than they fill out ballots, there just isn’t a perfect way to be sure you’re polling people who will actually vote.