Robert_Charlton - 11:24 pm on Aug 11, 2008 (gmt 0)

If I'm remembering correctly, KWD uses toolbar data.

Some of their data that involved placename information used to be skewed, I felt, suggesting that not enough of the toolbars were in the US... or at least not evenly distributed... but I believe they've said they've fixed this. Their numbers, btw, aren't searches per month... and I don't know whether they even try to predict that anywhere.

The numbers used to be just accumulated numbers of searches since they pressed the "go" button. They got their granularity simply by counting searches over time. It appears now that they've started the count over, and I haven't looked at the recent documentation thoroughly enough to say what that means, and whether the meaning of the numbers has changed as a result.

WT multiplies its very small sample size up to account for Google's very large market share. Doing this also multiplies the errors... maybe by a factor of 100 or more. I would NEVER trust WT for infrequently searched phrases, nor would I trust it as a predictor of the number of likely Google searches.

All of these tools suggest relative frequency better than they suggest absolute numbers.