Back in June, I detailed the sun setting on Google Website Optimizer in favor of the new Google AdWords Experiments. For the most part, I’ve been really happy with the new Experiments option. As I mentioned in my previous article, there are real benefits to the new approach, including the ease of setup and encouraging goal setting in Google Analytics.
Since that time, I’ve been using Experiments pretty extensively, and I’m still very happy with the platform overall. However, about a week ago, I encountered a small glitch that testers using Experiments should take into account.
While I was doing some site searching via Google using the “site:” operator for my own Web site, I found an alternative version of my home page -- one that was being used solely in an Experiments A|B test and had no links from other pages or sites pointing to it.
Why is it a problem if Google indexes and subsequently possibly ranks a test page in its index? It’s a problem because it will likely skew your test results, essentially giving you a false positive.
To figure this out, let’s start with the basics -- how Google Analytics Experiments works. The setup of Google Experiments only involves a few steps:
(To see the steps demonstrated, you can also watch a short video demonstration I recorded.)
Step two is the key step in Experiments. This code, which resides at the very top of the original page, tells Google Experiments when to serve page A (the original) versus page B (the test version). This code, however, does NOT appear on version B -- only on version A, meaning that the test is controlled through page A.
This means that visitors to page B will not truly be a part of the test -- they never were considered for both options A and B because they did not travel through page A and the testing code. It also means that visitors who come directly to page B can then potentially skew the test because they are adding to conversion data for goals -- even if they were not a part of the experiment itself.
While Google Website Optimizer worked in much the same way, Google did not index the test page automatically. And unfortunately, there is not currently an easy checkbox option in the test setup to disable indexing or apply a canonical redirect automatically via the test itself.
So what’s the answer? Google added some information in Google Analytics help to address the SEO impact. Their suggestion, which I recommend, is to apply the canonical tag to the test pages referencing the original page URL. Also, because these pages are actually indexed, it’s important to apply a 301 redirect after the test is over so that any traffic can be redirected to the appropriate test winner.