Here are our tips and tricks to running your own AdSense optimization. This is a mixture of methods, rules of thumbs, and suggestions from us for this very complex process. If you really want to extract all possible value from your AdSense ads, we naturally suggest you try out YieldBuild — it executes an extremely complex set of optimization tests algorithmically (also since a group of good layouts tends to outperform one great layout, and because optimization is an ongoing process) But if you’d like to do it manually, here’s what we suggest:
1. Review our advice on AdSense Formats and AdSense Placement
These will give you solid rules of thumb. However, remember there are always exceptions:
We know of one site, for instance, that has a blue-and-white background, where a bright green ad performs best above the fold, totally against the rule of thumb that suggests blending your above-the-fold ads into your background, or choosing subtle hue variations.
Another rule that is often broken is one of ad unit size. We suggest going with the 3 most popular ad unit sizes. However, if you’re on Blogger, for instance, the 728×90 ad size is not supported. You’ll have to go with 468×60 or 234×60. Similar deviations from the three most popular might have to be made for your site, based on your platform or template design.
So, use our suggestions, but try a wildcard option based on your intuition, to see if a deviation might actually perform better.
2. Pick a testing protocol
This can go from very simple (with likely less-impressive results) to more complex (with better results).
A very simple protocol might involve:
- choosing a very small sample of color/format variations, based on our rules of thumb, your intuition, and maybe one wildcard that you’d like to test (like a bright, constrasting set of colors)
- similarly, choosing a small set of layouts
- ignoring time of day effects (i.e. assuming traffic at different times of day perform the same)
- ignoring traffic source (i.e. assuming that visitors from any source, whether they’re loyal users or one-time visitors, behave the same)
- assuming no changes in your traffic month-to-month
Pick a certain time interval for each test phase that you feel would give you enough useful data, at week-long increments (less than one week increments are not recommended, since, as you probably know, traffic varies quite a bit over the course of a week).
Set up a reasonable set of permutations on format and layout (for instance: Color1 x Layout1, Color2 x Layout1, Color1 x Layout3, etc).
Try each variation of formats and layouts for each time interval, recording as much performance data as you can, but especially focusing on CTR (click-through rate). Assuming no significant variation among advertisers targeting your site or its content, and no substantial fluctuations in traffic, you should have some degree of information about which format/layout permutations work best from among those you tested.
For a more complex protocol, but that provides greater accuracy in your results (because it doesn’t assume traffic or advertiser activity stay the same, week after week), we suggest sequential A-B testing. In this case, two layouts with their corresponding tested formats “face off” against each other, automatically switching each page impression from one layout to the other, in order to capture 50% of impressions with one layout (“A”) and 50% of impressions with “B”. Here are some protocols you’ll want to check out:
- Performancing’s PHP-based A-B testing script (also note the first comment’s link to a downloadable PHP script)
- Webmaster Tutorials PHP-based A-B testing script (great explanation and PHP script)
- ClickZ’s A-B Testing for the Mathematically Disinclined (a good backgrounder on how A-B tests work, and how to ensure the proper sample size)
Two more; hat-tip to Danny Gabriner:
- SplitTester allows you to enter two ads’ clicks and CTRs and it will make a determination of the long-term performance; this is useful if you have a relatively low-traffic site and don’t want to wait until you reach statistical significance.
- Blazonry: AdSense testing PHP script
For a higher level of optimization rigor, you can invest a little time to design a Taguchi Multivariate Testing (MVT) or Design of Experiments (DOE/DOX) protocol that will allow you to specify the parameters, their ranges, and optimum outcomes. It will, in turn, generate the precise format/layout combinations for testing. What we haven’t been able to find, so far, are any scripts that Web publishers can use to run this sort of test on their own Web server.
3. Test, record, repeat
Apply your protocol, set up custom AdSense channels, determine the permutations that perform better, and…repeat.
Repetition is necessary for a few reasons:
- Traffic patterns can change. Fluctuations in the source of your site’s traffic can have an impact on how your audience responds to ads. Traffic changes can include changes in source sites, visitor geographies, and peak browsing time patterns.
- Advertiser patterns can change. Site-targeting advertisers can add or drop you, and there can be drifts in pricing or in campaigns for the keywords that your site targets.
- Ad blindness can set in. Especially if you have a site with a large number of repeat visitors, your readership can learn to ignore ads that don’t change position and color. Rotating among different ad formats and layouts, and occasionally jostling the formats with something new and even jarring, can help stave off viewers’ subconscious ability to ignore your ads.
Although we think this is both extremely complicated and extremely valuable (which is exactly why we designed YieldBuild), we know that some publishers like to take control over this process and reap substantial benefits to their AdSense revenue. For others that want the same (and better) performance without the time and hassle, we’d be happy to welcome you to our service.
Many thanks to Chris Gathright for contributing to this article.
This entry was posted on Wednesday, June 11th, 2008 at 10:33 pm and is filed under Online Advertising. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.