People who mistrust or discount surveys off hand generally do not understand how surveys work. Surveys, like anything else are on as good as the questions being asked and thesampling methodology employed. If the person generating the survey is biased and is making no attempt to look for neutral language to solicit responses, then of course its going to be biased.

The second thing to look for is what is known as a scientific survey. This does not mean they are asking scientific questions, rather they are sampling a random sample of the population. The randomness is what legitimizes the survey, along with the sample size. A good scientically-based survey controls for the language used in the questions so as not to lead the participant taking the survey in one ditection or another.

A good example; if you were to go to a Trump rally and ask participants about their perception of the Democrats, you would expect to get a unanimous negative answer. First off, the sampling population is already biased against, Democrats. It is not a random.sample. The outcome is already predictable as the Republicans feelings toward Democrats is already known.

If you however you do a random phone survey where numbers are generated by an algorithm, your chances are greatly improved that the trend of your survey is actually what it suggests (as long as your sample is statiscally significant).

Not all surveys are created equal. Sources like Quinnipiac take greats pains to be language neutral and scientific in their approach. CNN also, despite the oppositions misplaced charges also attempts to make their polls scientific. Unfortunately the same cannot be said of their competition. FOX news is well known for only polling the population they know will give them the opinion outcome they want, or in asking leading questions introducing bias at the outset.

You have to take surveys in context of how they gathered their sample. Sefl selecting samples are seldom a good measure of the general population.