The polls I pay attention to are those cited by the major news outlets. They identify all those variables, including the question, the polling organization, and the margin of error.

Professional pollsters take a stratified random sample of about 1,200 people. And, yes, that's enough to give a good ballpark estimate. The key is that the sample population be representative of the whole: random (everybody has an equal chance of being polled) and stratified (every subgroup of the population is covered). If properly conducted, the margin of error is about +/-3% 95% of the time.

Consider that polling in 2016 indicated Clinton would probably be elected. In fact, she did win the popular vote, which is what most of the polls were measuring.

As for the electoral vote, polling showed both candidates within the margin of error, i.e., the election could go either way. As it turned out, Trump won more than Clinton because of the geographical distribution of the votes.

Possibly the biggest caveat about polls nowadays is that people are now more likely to give false answers because they think the polls are rigged by the Fake News and the Deep State. Another is that caller ID keeps many people from answering the phone.

There are statistical methods that can minimize such bias in the sample.

A well conducted poll isn't infallible, but it's far from meaningless.
Even in 1948, when most polls had mistakenly predicted that Dewey would beat Truman, a swing of only about 1% in California, Ohio, and Illinois would have given him the victory in the electoral college.