Plenty of good reasons to doubt any method of trying to determine human thoughts. The fundamental flaw in understanding polls, however, is ignorance of the basic mechanics.
If you took the human element out and sampled, say, a dump truck bed full of red and blue gumballs, the limitations would clearly show.
I haven’t checked the numbers, but my guess is that presidential elections rarely finish worse than 55/45 in the popular vote. Well, with a MOE of 3.5, the MOE covers 70% of that range. Thus, when you see anyone say, “Well, that poll hit it right on the nose last time,” it betrays a total ignorance of what MOE means. They’re just mouthing a buzzword to impress you. ‘Hitting it on the nose’ is pure happenstance.
Conclusion: polls can give hints of trends, but are incapable of getting anything exact. Even with gumballs.
Bonus: no point within a given MOE is more likely than any other point in the range to be the actual result. A reported result of 43% +/-3 means 40 or 46% (or any point in between) are just as likely to be the correct answer.
If it wasn’t for vanities, threads straining at gnats over poll results would be the biggest waste of time on FR. There is no such thing as “he moved up 1 point today.”