Optimizing Google AdSense

Google's AdSense scheme has bucked the trend of declining Internet advertising. Web pages can easily show relevant adverts and get paid on a per-click basis. This has proved profitable for many information publishers.

However, it is very difficult for a web publisher to choose the best display options in order to maximize advertising returns. At first this was difficult due to the coarse nature of the statistics provided by Google, but with the advent of Channels it has been much easier to determine which areas of the website, and even individual pages are the most profitable.

It is more difficult to determine which layout for a specific page is the best. Some argue that a more prominent advert is best, whilst some argue a more subtle and subdued advert is best. A larger advert will hold more advertisers, increasing the chances that at least one will appeal to the reader. But smaller advert will intrude on the page less. All these arguments are valid, although some apply to to some sites more than others. Finding the exact mix is difficult without long and tedious trial and error.

This page demonstrates how you can quantitatively find the best AdSense layout for a single page. The second part includes a calculator to statistically calculate if two clickthrough rates are similar or different. Although unnecessary in "obvious" cases, this is very helpful for two adverts that appear to be producing similar clickthrough rates.

Testing Two AdSense Layouts on the Same Page

First, create two AdSense Channels. Next, create the two pieces of Javascript for the two AdSense layouts that you wish to use. Use Google's AdSense interface to set one layout to use one Channel, and the other layout to use the other Channel.

Combine these two pieces of code, using the following JavaScript template:

<script type="text/javascript"><!--
google_ad_client = "insert-your-client-code-here";
if (Math.random()>=0.5) {
   google_ad_width = 120; 
   google_ad_height = 600; 
   google_ad_format = "120x600_as"; 
   google_color_border = "578A24";
   google_color_bg = "FFFFFF";
   google_color_link = "00008B";
   google_color_url = "00008B";
   google_color_text = "000000";
   google_ad_channel ="insert-your-first-channel-code-here";
} else {
   google_ad_width = 120; 
   google_ad_height = 240; 
   google_ad_format = "120x240_as"; 
   google_color_border = "578A24";
   google_color_bg = "FFFFFF";
   google_color_link = "00008B";
   google_color_url = "00008B";
   google_color_text = "000000";
   google_ad_channel ="insert-your-second-channel-code-here";
}
//--></script>
<script type="text/javascript"
  src="http://pagead2.googlesyndication.com/pagead/show_ads.js">
</script>

Cut and paste the above code into your webpage. You must set the google_ad_client to be your client code, and the two google_ad_channel values to the codes for your two test channels. Set the other values according to what you want to use for the two layouts. This particular example has identical colors for both test channels, and only tests the total size of the AdSense frame.

The above Javascript randomly chooses one advert layout or the other. Try it.

Run the above code (with your layout settings) for a few weeks to get a good sample of which layout produces the best response. Sometimes this will be very clear, but you can use the calculator at the bottom of this page to test the statistical significance of any observed difference.

Are Two AdSense Click-Through Rates Statistically Different?

We have now collected some test statistics using Google's AdSense. These statistics might use my AdSense Layout Comparison Code (see above), or they might be from different pages or even sites. Either way, you have counts of the number of impressions and the number of clickthroughs for two different AdSense Channels. Sometimes these are easy to compare, but often they are not. This might be because one Channel recorded many more impressions that the other Channel. Or it might be because both the recorded impressions and clickthroughs are very similar for both Channels.

Use the following calculator to determine if two hit rates are statistically different or not:

Here are the results:

z (no. of standard deviations) is a standardised measure of how different the two clickthrough rates are. A value of zero indicates the clickthrough rates are statistically the same. A value above about 1.65 indicates the two rates are probably different.

The Minimum Significance Level is more useful though. If a value is given, then this is the probability that the two clickthrough rates are different. The lowest indicated level is that of 90% probability. If a value is reported as "similar", then there is either insufficient data, or the clickthrough rates are effectively the same.

If you are getting a null result, then it is difficult to say how many impressions are required before you stop. It is probably a good idea to run the test over at least a few weeks, so that you can include a typical range of variations in user behavior. A few thousand impressions over six weeks or so is probably sufficient.