Issues with geo targeting and reporting on Google Marketing Platform
One question that comes up again and again, especially among local business advertisers, is “I excluded city X from my Google Ads campaign. Why am I seeing website visits coming from Google Ad clicks when I check my campaign performance in Google Analytics. Often this question is not asked dispassionately, and for good reason. When the advertiser mentally agreed to pay for the click, he believed that his ad would be served only to people who met his targeting criterion. So it makes sense the advertiser would be less than thrilled to find out the money wasn’t spent in the way he felt he agreed to.
In this post I’ll attempt to explain a bit about how I believe geo targeting works from a technology standpoint. I’ll then try to drill down into advanced location settings and how Google Ads and Analytics specifically gather, utilize and report on user location data. You may have noticed that two sentences ago I wrote “I believe” instead of “I know” when referencing how user geographic data interacts with ad serving. One important thing for advertisers to understand is that we are not experts on what data points were used to determine the ad serving or what data points were used to report the user location in Google Analytics. Google does not make public really much of anything in the way of technical information about how it uses location data to influence ad serving or how it uses that same data in advertiser reporting within Google Analytics. That information is proprietary, so we only get vague public statements like this and this and this, which you can read if you want to review the information on this subject that is publicly available.
Think about it from the publisher side for a moment. Google search has literally milliseconds to collect all of the information that it can about both the buy side and the sell side of the search query interaction. The more time it spends collecting and analyzing data, the fewer people will use Google. In test after test Google found that two or three tenths of a second speed improvement on serving a search result page produced huge gains in user loyalty and acquisition. In this reality where milliseconds matter, I think it makes sense to use exact location data if you 1. have the necessary data, and 2. can charge a premium for using it. It makes sense to use less accurate location data if you don’t have the data or you aren’t getting a premium cost per click for using it. If we accept that Google is probably using profit maximizing algorithms that think along these lines, some of our questions become easier to answer.
Now, let’s outline a few of the things that an intelligent layperson who is willing to spend a little time reading stuff on Google can confidently know about how location targeting works in general. These data points and understandings will be useful when thinking probabilistically about how Google Ads and Google Analytics specifically use and report on user location data.
What we know about user location data
Gen 1 IP Addresses
I’d guess that at least 80% of the people who have ever asked a question like this already know that there’s this IP address thing and it is definitely a piece of the location targeting pie. It is, they are correct. Per Wikipedia “Each ISP or private network administrator assigns an IP address to each device connected to its network.” Per Google “IP addresses can often be used to identify the location from which a computer is connecting to the Internet. When you set location targeting for your ad campaign, IP addresses help Google Ads know which customers seem to be using a computer in your targeted region” (emphasis is mine) There are many cases in which IP address cannot be used to correctly identify user location. The most obvious one is when a user is accessing the web via a VPN, but location errors are common if the only method of determination used is IP address. Modern DSPs use endeavour to use multiple data points, like phone GPS, nearby WiFi and more, to determine user location. Explaining in more detail about this specifically is beyond the scope of this post, but you can easily find countless examples in the SERP of a query like this one.
Gen 2 Wifi
Clear statistics on wifi useage in the US are a little tough to come by, but in any case we know that it has grown exponentially over the last decade. During that time digital ad platforms have been using many collection methods, including those awesome Google street view cars, to detect when users are connected to which hot spot and learn where exactly that hot spot sits on the map. Google explicitly states “If a device is connected to a Wi-Fi network, we may detect the Wi-Fi network’s IP address to determine physical location.” and “Depending on a user’s location settings, we may be able to use a precise location for advertising, based on one of these sources of location data:” … “Wi-Fi: Accuracy should be similar to the access range of a typical Wi-Fi router.” There are a lot of ifs, mays, and shoulds in there which help answer the initial question in the post in and of themselves. In any case it is safe to assume that available data about the device’s relationship with nearby wifi hotspots influences Google’s ad targeting. Quotes from About targeting geographic locations in the Google Ads official support documentation.
Gen 3 GPS targeting on mobile
Using GPS data is a big step forward if you’d like to improve the accuracy of your best guess about where a user is located. The question is, has the user agreed to provide you with that information. The answer is, probably. If you have the facebook app installed, use google maps, uber, yelp or dozens or really any app that has enough users to make an “anonymized user data sharing” type deal, your users probably explicitly agreed to share their location data with you. The vast majority of users provide Google, Facebook and others with their GPS location. It is likely that the most sophisticated ad serving platforms use any real time location data points they have access to in addition to historical GPS data they have access to for any given device. This creates interesting differentials in data sets between different ad platforms. Let’s say platform A has access to all Google app services giving them maybe 60% or 70% real time access to device GPS in a market like the US. Then platform B has access to location data, but for a smattering of less used apps like games, second tier weather apps, and so on. Both ad platforms can truthfully claim that they provide geographic targeting based on gps location data, but while both will make incorrect decisions about current user location when serving some subset of their ads, one will be significantly fewer mistakes than the other.
What we don’t know about user location data
A lot. For instance, which percentage of Google ad impressions are served based on only IP address information compared to those which are served due to multiple recent app logins, WiFi and IP data points all being used at the same time? If IP address is the only data point accessible when processing an ad auction, what is the expected accuracy range? If two week old GPS data is used, how often is the user still in the same spot, or close enough to avoid violating an advertiser’s specified targeting? We don’t have real answers to any of these questions.
I asked why is Google wasting my money serving my ad to customers I can’t sell anything to and you won’t shut up about geek stuff, get to it!
Gotcha, sorry about that, of course. So, as a practical matter, and appreciating some of the nuance that’s involved in detecting and using customer location data in digital ad serving, there’s one thing we have to do first. Accept that even if we do everything right, and the person we are buying the ad from is giving their best effort, unless we sell worldwide some percentage of our ad spend will be wasted due to bad geographic targeting. It’s reality and throwing a tantrum about it won’t help. Having accepted that, is there anything an advertiser can do to limit wasted ad spend? Sure is
1. Advanced Settings – Users in my targeted location
Here’s what it looks like to use the setting which will give you the highest accuracy in terms of only serving your ads in the areas in which you’d expect your ad to be served. Let’s say that your campaign is targeting Omaha, like this one is. Using this setting will let Google know that you would prefer that they not serve your ad when someone in, say Chicago, searches “YOURKEYWORD Omaha”. Using the default option will tell Google “please serve my ad to the guy in Chicago if he includes my city name in his search”.
2. Use location exclusions
To a certain degree this is going to feel a lot like
but I promise it’s worth it. For starters, I like to exclude all the U.S. states I don’t do business in. I also like to exclude Canada and Mexico, unless of course I do business there. If the client only does business in a small portion of the state, excluding cities and counties can help. For some reason, even campaigns that target a specific U.S. city will occasionally drive traffic that either Google Ads or Google Analytics identifies as out of country. In my opinion, the best way to minimize this is to explicitly exclude all non-US countries. There are only 195 of them. A few of them you can’t explicitly exclude, but don’t worry, they are blocked whether you like it or not, due to international trade embargoes. Let me know in the comments if you also know which ones these are, fun easter egg if you’re a search dork. The moral of the story is that if you find a location in either Google Ads or Google Analytics that is driving a lot of unprofitable traffic, the best you can do is exclude it.
I’ve done all that stuff and I’m still paying for out of market clicks
Gotcha, that sucks. For starters, remember we already agreed that location targeting and reporting aren’t perfect. It could be that either the serve side or the report side just happens to be incorrect. Everybody has to set their own tolerance as far as what percentage of spend is acceptable to lose to these inaccuracies. Mine is about 5%. If yours is a lot lower than that you probably have unreasonable expectations. Regardless, here’s my best answer to a recent questions from on the Google Advertiser Community.
…”I am targeting my ADs in in a 20 Km radius from Sydney CBD, Australia.
At the same time, i have also entered few Excluded locations throughout Australia eg. state of Victoria, state of Queensland, Western Australia, etc.
Under “advanced location options” set – the default settings are selected (Target: People in, or who show interest in your targeted locations and Exclude: People in, or who show interest in your excluded locations)
Issue is that my ADs are being seen and clicked in Melbourne (Victoria), Brisbane (Queensland), etc.”…
Google Ads determines user location with a waterfall type analysis. Since it has to match ads to searches in milliseconds, sometimes Google Ads only gets a very high level geo area before having to push the ad. Sometimes Google knows where you are down to the square meter, sometimes it can only tell what country you are in. Depends on device, browser, app, signed in status, using WiFi, how long it’s been since a Google streetview car drove by where you happen to be and probably a hunderd factors Google doesn’t tell anybody about because they don’t want Bing to start using it to improve their results. Google saw the search included a term for your targeted region, didn’t know for certain that the searcher was in your excluded area yet and served the ad. After the user clicked, and spent some time on your site Analytics was able to find out that the user was in Melborne or Brisbane. Not having a time machine, Google can’t go back in time and decide not to serve the ad.
Location targeting isn’t perfect, but there are things you can do with Google Ads, Google Analytics, and other platforms to improve the percentage of your web traffic that comes for the location in which you can do business. Overall, this is a complex subject and is constantly changing and being innovated on by publishers. As advertisers it is our responsibility to understand what is within our control and optimize that while understanding and monitoring for an acceptable failure rate that balances cost of errors and relationship with the publisher. Happy pay per clicking locally!