Kissmetrics Blog

A blog about analytics, marketing and testing

Built to optimize growth. Track, analyze and engage to get more customers.

Can Split Testing Wreck Your Google Ranking?

Put down your pitchforks and torches… I’m not saying you should stop split testing. But I am saying that there are a few “red flags” that a split test can inadvertently throw up if you’re not managing or monitoring it correctly.

It’s a fine line you have to walk if you do both conversion optimization and search engine optimization. Most people think the two can’t co-exist peacefully, since you’re either writing for people, or you’re writing for robots. But you have to ask yourself – what’s more important? Convincing customers to order or appeasing an algorithm?

Let’s take a closer look at some common split testing mistakes that can cause issues with your ranking – and how to fix them.

Deleting the Under-performing Test Images

Let’s say you’re selling the latest and greatest smartphone – and you’ve got a deal going that not even your biggest competitor can match. You do a split test on the “deal of a lifetime” page showcasing two different views of the phone – one standalone, and one being used by a consumer. As it turns out, the one with the consumer wins.

You then do what most folks would reasonably do after a test – you delete that underperforming test page with the standalone image.

STOP!

While it’s completely natural to want to delete all the things that make up the failed page (and erase any evidence of its poor performance), you may be missing out on some visitors who come to your page solely through the image (or video) itself. Nothing is more frustrating than wanting to see a close-up or watch a video in higher resolution only to get a 404 error – and Google doesn’t like it either.

Depending on how much traffic you get, it could funnel hundreds of visitors somewhere else instead of your pages. Of course, you can remove the actual challenger pages that you created with these images or videos, and simply 301 redirect them back to your winning page – the search engines are totally okay with that.

Optionally, you could overwrite the poorer performing image with the winning image and save it under the same file name. No harm done.

Filtering Spiders from Your Tests

Some split testing programs give you the option to filter out spiders from your testing. This is done via a little piece of code called the user-agent. The problem is, Google wants to see exactly what the user sees, not some doorway page that’s optimized “just” for them. Although this practice was a big deal a decade or so ago, cloaking content seems to have gone the way of the dinosaurs, and thankfully so.

Still, if you’re using older tools or anything that allows you to filter or redirect spiders like Googlebot to another page – just say no. It’s a dangerous path that can lead you straight into demotion territory.

But I Don’t Want Google to See My Test Pages!

This is a common issue with split testing – you want to let Google in, but you don’t want them to mistakenly index the wrong thing. You can check to see if Google is allowed in the first place by looking at your robots.txt file. This file gives general guidelines to search engine spiders on what they can and cannot access.

Some crawlers will abide by these guidelines (Googlebot will), and others won’t. But even though Google won’t index those pages on your site, they might still find and follow those URLs from other sites. In these cases, Google suggests alternative methods of blocking beyond the robots.txt file, such as password protection.

Fortunately, you can check to see if your test pages can be found in Google by typing this into the search bar:

site:example.com/testpagename/

Replace example.com with your URL and testpagename with the name of your test page. If you see your page in Google’s results, then you’ve been indexed. You can add a snippet of text to your HTML <head> that reads:

<meta name=”robots” content=”noindex”>

And you can add this snippet to your robots.txt file:

User-Agent: Googlebot
Disallow: /testpagename/index.html

(choose the correct directory and filename for your particular test).

If you’ve verified the site using Webmaster Tools, you can also remove it from Google there. Be sure to do this only if you’re certain it won’t conflict with any of your existing measurement or analytics tools.

Prevent Duplicate Content Issues

Another common problem that can sabotage your search engine ranking is the dreaded duplicate content penalty. From Google’s point of view, there’s no sense in showing two of the same page if the only difference is a new headline or image.

moz

Source: Moz.com

The good news is that you can tell the search engine which version of your page to index by way of the rel=canonical tag. This is much faster to implement than your typical 301 redirect and bypasses the common glitch of creating an infinite loop when you direct example.com/index.php back to example.com

Here’s what to do while the test is running:

  1. Determine if you want your control or your variant (challenger) page to be indexed.
  2. Add the canonical tag to the page you don’t want indexed by the search engines using a link that points to the page that you DO WANT indexed.
  3. You can also add a NO INDEX tag to the head of the page you don’t want indexed.

Here’s the code for a canonical tag:

<link rel=”canonical” href=”http://www.example.com/mycontrolpage” />

(Replacing the URL with your own site and test page name, obviously!)

Without getting too technical, this snippet of code tells Google which version of your page you’d prefer them to index, and most of the time, it will index that one and go about its merry ranking way.

canonical

In this case we want the search engines to index our Control webpage

Most times you would probably want to have your control to be indexed because that’s the URL that’s going to stay put for the duration of your website’s life. However, each case is different, so think about it carefully.

If you’re finished testing and you’ve determined a winner, then you’ll want to:

  1. Update the control URL with the winning content if the control lost. If the control won, then leave it alone.
  2. Delete (or remove) the variant URL from your site.

Many testing software applications are good about using the same URL for both the control and the challenger page during the test. If your testing software does this, then you don’t need to worry about using canonical tags.

Here are two great articles on this issue from both Google and Optimizely:

The Bottom Line on Split Testing and SEO

Split testing and optimization can live together harmoniously if you manage, plan and organize your tests well, and take steps to remedy issues as they happen. Being proactive about your testing will not only help you discover winning variations to implement, but will also keep Google happy by consistently providing your users with relevant information – a win-win for everyone!

Has split testing affected your ranking in Google or another search engine? How did you deal with it? Are you finding other split testing issues as you try to improve your site? Share your thoughts and comments with us below!

About the Author: Sherice Jacob helps businesses improve website design and increase conversions with user-focused design, compelling copywriting and smart analytics. Learn more at iElectrify and get your free conversion checklist and web copy tune-up. Follow @sherice on Twitter, LinkedIn or Google+ for more articles like this!

  1. Interesting. I’ve been reading more and more about “link rel = canonical”. This helps in my becoming more familiar with how it works. Thanks!

    Sandy

  2. Michael Paulse Nov 26, 2014 at 8:26 am

    The post reads, “2. Add the canonical tag to the page you don’t want indexed by the search engines using a link that points to the page that you DO WANT indexed.”

    But I don’t see that in the image.

  3. Darragh McCurragh Nov 27, 2014 at 12:53 pm

    This was actually something I also had been wondering. After all, split testing is in theory “duplicate” content, depending on how you go about it or what you need to test. Or it is not, then you may distort PageRank if the A/B pages e.g. have different html addresses and/or get linked or shared via social media differently (although that in itself would spek volumes about the quality too, but it would obfuscate the SERP question you tried to answer in the first place). Thanks, Sherice, for sharing your insights!

3 comments

Please use your real name and a corresponding social media profile when commenting. Otherwise, your comment may be deleted.

← Previous ArticleNext Article →