Posted by Sam Battin, Senior Natural Search Specialist
Performics offers advice on multivariate and A/B testing of Web pages to improve conversions and sales while maintaining your SEO edge
Testing Web pages is very useful when building a successful website. Rearranging content, images, button locations, and other visual elements on your landing page can tell you which combination will drive the most conversions from your visitors.
At the same time, however, you definitely don’t want to run afoul of search engine guidelines. Is testing cloaking? Can it create duplicate content? Will multiple versions of your test pages get indexed? Web developers who want to maintain their current ranks while conducting tests stay up late at night thinking about these things.
So to help Web masters create test pages that are SEO friendly, Performics has developed the following guidelines to consider when using test pages. These recommendations have been built out from Google’s own advice about testing and the findings of Performics’ own Landing Page Optimization Team.
1. Test pages should uphold the “spirit” of your original content
Google provided the following statement on acceptable contents of test pages:
Your variations should uphold the ‘spirit’ of your original page’s content — they shouldn’t change its meaning or people’s general perception of it.
Generally speaking, a test page will only change certain elements present on the original page, such as images, the location or color of a button, or the text copy explaining the value proposition. While different versions of your page can convey slightly different versions of your message, the essential meaning of the content should be consistent across all test pages.
For example, if one version of a test page presents information about a free video rental and the second version of a test page contains information about a very different offer, such as a free video game or free membership, Google may consider the test page as “cloaking” because the meaning of the page, and visitors’ perception of its contents, have substantially changed.
2. Have a firm start and end date for testing
Google provided the following guideline for the duration of a test:
Keep your source code updated. An experiment should be ended when sufficient data has been collected, and the original page source updated accordingly.
Once the test has been concluded, the alternate version of the URL should be removed and Google should find only one set of contents at the URL. The implication of this guideline is that while Google will not penalize a page for testing, the test must be of a limited duration. Though Google does not specifically state a date after which testing must be concluded, it does say that “If [Google finds] … a site running a single non-original combination at 100% for a number of months … we may remove that site from our index.”
3. Show your original page to a non-trivial percentage of visitors
Google’s guideline regarding original pages in testing specifically states the following:
Your original page should always be shown to a non-trivial percentage of your users. It’s okay to use Website Optimizer to briefly run your winning combination after stopping an experiment, but be sure to update your test page’s source code to reflect that winning combination soon afterwards.
This refers to what should happen at the end of a test when the most successful combination has been found. A brief period of time to confirm the effectiveness of the winning combination is acceptable, but the original page should still appear to visitors at a non-trivial rate, e.g. not zero.
4. Test pages should include a rel=”canonical” attribute and a “noindex” META robots instruction
In cases where a test page redirects visitors to different versions of the URL, the test URL should not appear in the search engine index. For example, suppose the URLs for the original and test versions of a page were as follows:
The test-offer.html page should contain a rel=”canonical” attribute that names the original version of the page as the preferred version of the URL, e.g.
<link rel=”canonical” href=”http://www.yoursite.com/offer.html” />
This tells search engines to apply any inbound link value to the original version of the page, and not the test version.
In addition, the page should also contain a META robots instruction that states that its contents should not appear in search engine result pages, e.g.
<META NAME=”ROBOTS” CONTENT=”NOINDEX, FOLLOW”>
In this way, the test version of the page will not appear as a potentially confusing search result alongside the original version of the page.
5. Have a continual testing plan that outlines next steps
For best results when conducting a test, we recommend implementing a continual testing plan that clearly states what to do next when either the control page or the challenger page wins.
For example, if the control page wins, then the next test slated to run should immediately begin. If the challenger page wins, the process to implement the successful changes to the original should be already in place. This will allow the page to start obtaining higher conversions as soon as possible. In addition, the new test (e.g. the new challenger page) should be ready to test against the latest approved version.
Try to incorporate the above information into your testing process to make sure that search engines don’t penalize you for duplicate content or rank your test pages above the original URL. If you’d like some professional help in designing multivariate tests for your website, please contact Performics’ Landing Page Optimization team.