It seems to me that if there were one data source I would expect to not be consistent with Benford's law it would be polling data. Benford's law is usually seen in numbers that follow some skewed distribution. Polling data groups around maybe 45% depending on the number undecided. I don't know what I would expect, but polling extremely close races, or polling close to the election, I would expect to see lots of high digits (6 7 8 9). I wouldn't expect many 41 or 31 (those questions are uninteresting) so the 1s come from mostly 51s.

He may be right, but I'm going to need a little more convincing if someone wants to assume that polling data would exhibit Benford's law.