You're technically correct, but it is commonplace to say things like "60% of people want to repeal the Eight" or "52% of people would vote Remain on Brexit today" when what is really meant is "52% of people in a study/survey of 1500 people would vote remain on Brexit today". Nobody thinks that in these cases all people eligible to vote have been asked and chased until they give an answer to come to an result that is guaranteed to represent each and every person, we understand that some smaller number has been used to give an estimation of the view of the whole. We accept it won't be perfect, so won't read much into a survey that is tight like 51%:49%. Again I'll ask what would you suggest as an alternative here? It's not like they could hold a referendum on it, nobody would bother to show up. An outbound randomised sampling would likely be a more accurate and less susceptible to brigading or selection biases, but it's very unlikely to turn an 84%:16% result around. I'd also imagine that an inbound survey like this is a good way to test the waters of interest, which can then be followed up with an outbound randomised survey to improve the certainty of the data.