Almost never. Too many organizations doing polls that have no business doing them.
Creating a non-biased poll is part science and a lot of art. You have to create questions that are clearly understood by the vast majority of those that you poll, you then need to be consistent in the coding of the answers, and correctly calculate the results.
Now, to make it more complicated, you need to create a panel. A true random panel of the population you want to measure is the most valid but almost no one goes to the trouble of attempting one. The other more valid alternative would be to create a sample that is representative (in terms of demographics, opinions, etc) of the population as a whole. This method is not ideal but is better than what you typically get which is a narrowly focused self selected sample with a lot of biases.
Most polls contain misleading (or outright biased) questions, large gaps in both sample size and representation, and the assumptions that the publishers make based on the "results" are more often than not leaps of faith with no real statistical footing.
But other than that, they're great.
