Testing the accuracy of Google Insights data


Obviously, getting any useful information about search volumes from Google has always been a bit like getting blood from a stone. Robin had a look at what you can do with the Google Insights tool and since then we’ve been quite intrigued to find out how accurate Google Insights actually is.

We arranged a little test to see if we could translate Google’s search volume ratios into actual numbers.

So we took a relatively high volume financial search term that we currently bid on for one of our clients. The client has enough budget to ensure the ad shows on every search, and the ad always appears in the top position. Then we graphed the number of impressions for this keyword over the last year.

We then took the data for the same keyword, across the same time period from Google Insights and mapped them on top of the impression count from Adwords.

Initially we could see the same peaks and troughs, but the lines didn’t overlap on the graph. The troughs and peaks were in the right place, so it seemed as though the numbers were just out by a fixed factor all the way along.

To combat this, we introduced a fixed coefficient to the numbers from Google Insights, and sure enough, after a few calculated guesses, the lines soon started to overlap quite nicely.

Image of search volume - Google insight vs Google Adwords

You’ll see though that the phrase match line was still slightly lower than the rest. We amended our calculations so broad and phrase match could have different coefficients. With a few more calculations we ended up with this:

Image of search volume - Google insight vs Google Adwords

These seem fairly accurate as far as we can see. So, we’re happy to say that both of the coefficients we found from this are between 50 and 80.

Bear in mind, if we could find this coefficient, we would have the best keyword tool on the web with data coming straight from the horse’s mouth!

Have you tried any experiments like this? If you have, tell us if you’ve managed to get similar results, and what coefficients you’ve found. If you haven’t and you thought this post was interesting… snap to it and tell us how you get on!

More on this subject