I've never heard of it before, but I did find this: http://support.google.com/adplanner/bin/answer.py?hl=en&...

I'm sure the process is fraught with errors, of course, but there does seem to be data to feed the model and data to serve as a test set. The easy thing about targeting ads with demographic data like this (unlike trying to make statistically-accurate statements about demographic data like this), is that if a person responds to some types of ads like a 45-year-old scuba instructor or whatever, it doesn't matter if they really are any of those things.

Well lets check result of their methodology with other data. For example for Slashdot. Geeknet provides very different data http://geek.net/our-network/slashdot/ Linked article suggest almost 50% in age 35-54 but geeknet stats says it's only 5%. I cannot find it now but someone wrote that about 1/4 or 1/3 of slashdot audience are female which again differs from result in article. I looks like that either US audience is very different from global(although it seems unlikely on US centric site like Slashdot) or methodology of linked article indeed is far from perfect. Obviously I checked only one site from article and there are other possibilities (like geeknet lying or having bad methodology itself), but I would take pingdom report with a big grain of salt.