I think there is a large element of truth to this. Especially if you can hand pick your demographics/region. A pollster in Texas would be a bit different than a pollster in Cali.
For the 'professional' pollsters covering politics they tend not to be deliberately biased as their reputation is built on being accurate. The reason they are not accurate is due to the difficulty of constructing a representative sample. This is why the margin of error on most polls should really be doubled from the stated level.
All of the statistics are built around having a random sample, but because polls are 'opt in' they are not truly random. The type of people who are willing to participate may often don't reflect the total public. Also people may be less willing to give certain answers is they are perceived to be controversial. Trump polled higher in automated polls rather than those carried out by a real person for example.
In addition, even given exactly the same polling data, different pollsters can arrive at different results due to methodology, while both are using perfectly legitimate methodologies.
Language is also very important and this is often better than sample composition if you do want to manipulate a poll result (it also has a big effect when people are trying to be objective too). You can genuinely turn a 75%-25% in favour in to a 75-25% against by rewording a question in a way that still means pretty much the same thing. For example:
Do you believe that those who can afford it should pay tax when bequething their estate to family members?
Do you believe that people should pay a death tax when passing on inheritance to their bereived family members?