Had another conversation with AdSense folks the other day.
Note: it is amusing that they don't know, at least at this level, that I operate multiple sites in the niche.
Which wouldn't have been note worthy except that earlpearl made a thread comment that resonated:
Google has mega data. We have little data. Google aggregates all its data about clicks of this sort; people searching for dentists and doctors, and restaurants and hair salons etc. They search in their own cities. They search in other cities.
Now, the above quote was in regard to how G might decide whether a query is local. I'm quite happily taking it out of that context and reapplying it in the resonating context: why sometimes AdSense advice seems wonky even contra-indicated.
Because they are looking at the big picture and not my piece of the picture. Just because a forest is primarily conifers does not mean that pockets of deciduous don't flourish. However, in the big picture they, if not disappear, diminish to insignificance.
And that is one of the critical problems of big data: one size does not fit all.
* surprisingly easily gamed
* increased risk of spurious correlations
* often merges data collected variously for differing purposes skewing results
* can reinforce error
Of course proper methodologies minimise all the above. However, the one most likely to not lose the trees in the forest, running multiple tests against multiple subsets, is not common. At all.
So, when AdSense 'advisors' make suggestions they are mostly/always based on the big data for the entire niche. And so they make, in amongst the beneficial suggestions, unhelpful even disastrous recommendations. The old saw about not extrapolating the specific to the general works the other way as well.
Take their advice should they proffer it and apply it against your own analysis of your own metrics. Only if it appears to fit should you decide how best to test it.
Regardless, always be pleasant and say thank you. For they may be offering a zillion dollar opportunity.