There’s actually a few different methods that can give you at least more accurate results if not 100% accurate (which polling never really is in the first place.)
Eg. list experiments are a potentially useful method. You start off with a list of statements like “I like candy” or whatever and you ask people how many of those they agree with (ie. not which ones, just the amount), which gives you an approximate baseline. Then you give another set of people the same list but with eg. “I support the war in Ukraine” added on (hypothetical example, nobody please get pedantic about the wording), and you then compare the total number of agreed-on statements with your baseline. Here’s an example from LSE last year that used this method: …lse.ac.uk/…/do-russians-tell-the-truth-when-they…