I know the polling agencies do micro-polls on the grammar before conducting the full-scale poll.

Gonzo journalism is unavoidable.

You don't want to ask a shit question, but that means you have to tweak your question based on the answers.
Good polling agencies do this well, but it doesn't change the nature of the problem.

Sometimes you will see a polling result that shows difference in answers based on the subtle change in the way the question is asked. That is presented when the difference is more interesting than either result on its own.