How accurate are google keyword estimates for local search volume?

[aj613]

We've all used the Google Adwords Keywords Tool, and if you're like me you use it to analyze data for a particular region.  Does anyone know how accurate this data is?  For example, I'd like to know how often people in Savannah, Georgia search for the word "forklift".  I figure that Google can give me two kinds of data when I ask for how many people in Savannah search for "forklift".

1. They might actually give me rough data for how many people in the region actually searched for the term "forklift" over the last 12 months, then divide by 12 to give me a monthly average.
2. Or they might use data on a much broader region and then adjust for Savannah's population size.  In other words, they might say, in the US people searched for "forklift" and average of 1,000,000 times a month.  The US has a population of 300,000,000.  Savannah has a population of about 250,000.  250,000 / 300,000,000 is 0.00083.  1,000,000 times 0.00083 is 208.  So, "forklift" is searched in Savannah an average of 208 times.

1. is obviously much more accurate.  I suspect that 2. is the model that Google is actually using. Does anyone know with reasonable certainty which it is?

Thanks,
Adam

[MiriamEllis]

Hey Adam,

Honestly, I am not able to answer your question with total certainty. What I've personally experienced is that Google's volumes for local keywords have always, always felt "off" to me. When a tool tells you that no one in X city is searching for something you are positive they are searching for (like water heater repair, or housekeeping, or yard maintenance), you're just left saying, "huh?". I thought this recent Whiteboard Friday did a good job of explaining why some of Google's estimates aren't reliable, in terms of Adwords:

https://moz.com/blog/unreliable-google-adwords-keyword-volume

I'm hoping another community member will be able to give you a more satisfactory answer on this, but thought you might like to watch that video in the meantime.

[Smileworks_Liverpool]

Hey Adam,

That's so funny because I use your basic model to predict my local volumes too.  Most of the stuff I do is local and because it's also pretty new or technical Like 'Plexr Plasma Surgery' or ' Vaser Liposuction' etc, it can be a real struggle because sometimes it's not even there in the indexed words from the big tools.

But you can augment your data with more sources.  For example looking inside search console at the amount of traffic a certain keyword is getting and then looking at the average position CTR's for that particular word.  This is a very rough guide.  But then you can add another data dimension which is looking at the the national rankings for more obscure keywords with your favourite competitive analysis tool and then figure out almost what the 'total search' is for that word and divvy it up using rough data from Google trends for your location.

Google trends is your friend because If I can see that 'Botox' has a certain amount of national volume (probably pretty accurate) then I can compare it to 'facial filler' and model the two because people who have heard of Botox will also have heard of filler and are in the rough same demographic.

So it's all guesswork but educated guesswork.  And I recently predicted traffic for an 'Icon' white spot treatment page this way.  (this gets the patchy flouride spots off your teeth after braces.)  I made my page and used the main Icon website visitors from SEM Rush - minus dentists to give me customers actually searching and then used google trends to figure out how much of that pie Liverpool was going to get as opposed to London or the rest of the world and the results are pretty accurate.  Like spookily accurate.

I'm certainly not a mathematician but I can use a calculator and my gut feelings.  I think It's from forecasting in business (we're really good at that now) that's given me some ability to forecast in other areas too.  You just need to verify the datasets from as many independent sources as possible and discount or give extra credence to certain ones over others using your intelligence.

But the best way is to build a page (and because it's a random of low vol keyword you can rank 1-3 overnight and see how many visitors you get.  Shoot first ask questions later.  What's the worst that can happen?  You made a nice page for your visitors to read.  Some of my best performing content I got this way.