What exactly is SEO Split testing?
Before getting into the ways it can be done wrong (and correctly) it’s worth taking a moment for a moment to clarify what split testing in SEO actually is.
The CRO test is the most obvious basis for comparison. For a CRO-based test you’ll typically be conducting a comparison between a different version and control of a page (or groups of webpages) to see which one performs more effectively in the area of conversion. It’s done by placing your visitors to different categories, and then the buckets are shown with a variant version your website.
For SEO split-testing, we’re trying determine the version of a website will be more effective in the realm of search engine traffic. If we take a CRO-like method of dividing users, we wouldn’t be able to evaluate the results, since there is only one version of Googlebot that will only see one page.
To circumvent this issue, SEO split-testing using buckets rather. It is possible to take a portion from a site in which all pages are based on a similar structure (for example , the product pages of the eCommerce website) Then, we create a modification to only half of the pages in the segment (for everyone). In this way, we can determine the impact on traffic from the changes across the different pages, and compare it to a forecast that is based on how well the pages that control the traffic.
Common SEO Split Testing Mistakes
1. Don’t leave split tests running long enough
As SEOs, we are aware that it takes a time for the modifications we make to show up in ranking. If we run the SEO split-test, the results are borne out in the results. Like you will see in the graph below that it takes a few weeks or so for the different pages (in the black) to begin to surpass the predictions based on page that controls the traffic (in blue).
It’s tempting to be worried after a few days or so, thinking to think that our test might not have made a significant difference, and to simply write it off because of a unconvincing outcome. But, we’ve observed repeatedly that the results can be altered after a few weeks and you shouldn’t be calling it too quickly!
The other thing to keep in the present is the length of time that you are left to leave it following this initial flat phase the greater chance to be substantial and you’ll be able to have greater confidence in the outcome you discover.
An important note for those reading with a background of CRO I’m sure you’re screaming at your screen that you shouldn’t allow a test running for a longer period of time in order to reach a certain point and that you need to decide in advance when your test will end in order for your outcomes to count. It’s true in the case of a CRO test that is analyzed using traditional statistical models. For SEO split test, we determine the significance of a test with Bayesian statistical methods so it’s a good idea to continue to keep a test running until it is of sufficient significance and you’re able to be sure in the results at this time.
2. Testing pages which don’t receive enough visitors (or are controlled by a very small amount of pages)
The websites we’ve had the opportunity to conduct split tests with Distilled ODN have varied in the amount of traffic and so have the sections of sites where we’ve attempted test split tests. In our experiences using SEO split-testing, we’ve formulated a general rule of thumb that says: If a site section that is similar to pages isn’t receiving at least one hundred organic sessions every day in the total, it’s likely be extremely difficult to determine any increase due to your split test. If you’re receiving less traffic than those to the pages that you’re testing, any indication that indicates a result, whether positive or negative outcome would be diluted by the degree of uncertainty.
Over 1,000 sessions every days, in general, the greater the amount of traffic you’ve got the less of an upward pressure you can observe. The most small effect size we’ve been able to quantify with statistical certainty is a tiny fraction of a percent.
In addition to having a a significant amount of traffic in your website section, you should ensure that the traffic is evenly distributed over a vast number of pages. In the event that more than 50%% of your site’s organic traffic is directed to three or more pages, that means your test is susceptible to changes in the performance of these pages which has nothing to directly relate to the test. This might modify you to believe that the change you’re testing is having an effect , even though it’s really influenced by an unrelated aspect. With the traffic being evenly spread across the entire site section, you are ensuring that these fluctuations specific to the page be even and you’ll be more certain that any impact you observe is real.
3. Pages are tacked arbitrarily
When conducting CRO tests, the most effective method is to place each user at random to either the control or different groups. This will ensure each group is identical due to the high number of participants that are likely to be in.
In the SEO split test, it is necessary to add an additional level of nuance to this strategy. If a website section has a huge amount of pages, and where traffic is evenly dispersed over them, the random approach might determined towards a fair bucketing. However, the majority of websites have pages that receive more traffic as well as others that receive less. Additionally certain pages could be different in their trends and have surges in traffic, particularly in the case of a specific seasonal function.
To make sure that the control groups and different pages of a page are statistically comparable We design the pages in the such a manner that they:
- Similar levels of total traffic
- Similar distributions of traffic among pages within them
- Similar patterns in the flow of traffic over the course of
- Similarity in a number of other statistics
5. Testing pre/post-tests instead of A/B tests.
If people refer to it as SEO testing, they are referring to making a alteration to a particular page (or across the entire website) and then determining if the site’s traffic or rankings increase. It’s not a divided test. If you’re making a modification and watching the results, your analysis is subject to external influences, such as:
- Seasonal changes
- Algorithm changes
- Competitor activities
- Your website may gain or lose backlinks
- Any other modifications you make to your website during this period
One way of be able to determine whether a modification has an impact is to perform a appropriate split testwhich is why we came up with ODN. ODN in the first in the first. To account for these external factors it’s crucial to have a control set of pages that you can predict the outcomes of your pages that you’re changing and ensure that the change you’re making is creating an impact.