I’ve done copywriting, website optimization, SEO, and split testing for eight years before Google’s now-dead Website Optimizer was born. The main question I’ve heard about SEO and copywriting is: “Do you write for search engines or conversions?”
I ask you this question: “Is the page you’re optimizing core to your marketing funnel?” If it is a home page, email capture page, or your sales letter, you had better write for conversions. Destroying conversions by 80% from a bad headline does not make up for the extra 10 visitors a month you get from a minor on-page change.
When you focus a page on conversions, it’s vital you split test. This way you can get more sales by understanding the core of what matters to your customer base. However, you don’t have to throw out the baby with the water. SEO can still be done.
Rumors exist that split testing hurts SEO. It can according to Google.
You hurt either SEO or your business if you make the following five split testing mistakes:
1. Presenting version A to Google and B to Billy
Your website can tell if a visitor is a Google spider or little Billy on his home computer with what’s called a “user-agent”. Through this, you can easily present different information to different users. Google wants to know what young Billy will see when he does homework.
Some tools allow you to filter out spiders from tests. This can help achieve cleaner test data. It is a dangerous SEO practice to present version A to Google and versions B or C to little Billy. This technique is known as “cloaking”.
Seen Mission Impossible 3? Tom Cruise’s character Ethan Hunt uses masks to “cloak” his face. It confuses the heck out of other movie characters… and viewers! Don’t be Ethan to Google.
I know Visual Website Optimizer is a popular split testing tool that does not cloak content. ZenTester is another good tool that gets a green tick. Your tool should not present a unique page to Google.
Tools today seem okay with this best practice. I do not know any testing tool that cloaks, but be safe. Ask your tool’s provider if they cloak content.
2. Allowing pages made solely for testing to be indexed
You don’t want test pages to be indexed because your control (by definition) is your best version to-date. Some split test tools do not redirect visitors who land directly on a temporary test page so you may hurt conversions.
Check what split-tested pages are indexed by searching each page in Google with the following query:
site:yourdomain.com/testpage.php
Does the page show up? Google has indexed it. Before you freak out, there are some tools and best practices to handle this perceived emergency:
1) Use your robots.txt file to block Google from the scanned page you’re testing. Add the following to your robots.txt file:
User-Agent: Googlebot
Disallow: /products/product-a-my-test-page.php
If I’m doing competitor research and you seem switched on, I’ll look through your source code or javascript snippets to see if you’re doing a split test. I’ll then search your robots.txt file to see if you’re telling Google to not index a page. Bam, I can see what you’re split testing then gauge valuable information and ideas.
If this scares you, befriend me or use method two.
2) Use the noindex meta tag on a page you want Google to overlook. Google generally drops a noindex page from its search results. Copy-and-paste the following between the
Never add this to a template file because Google will not index all pages that use the template.
3) For a solution to remove a page quickly from Google’s search, log into Google’s Webmaster Tools. Click the “Remove URLs” tool under the “Optimization” tab then create a new removal request:
3. Not using rel=”canonical” on tested pages
An A/B test of different URLs can cause duplicate content issues. Google will drop one of your pages out of its index because of their similarity. This may mean your original page gets removed from search results. Look at it from Google’s perspective: why show two pages that only have a different headline? It clutters search results.
Open the URL of a page you’re split testing. Copy a string of five or more words that are the same on all versions of your tested pages. Fire up ye olde Google.com in a new tab. In quotation marks, paste the series of words then search. What shows up?
If a tested page appears, in my experience Google will show this over the original page for most search queries.
There’s a solution! Canonicalization.
Use the code I’m about to give you on A/B tests where you test pagea.php versus pageb.php. Make the control page the canonical URL. So for pagea.php AND pageb.php, between the
http://www.yourdomain.com/controlpage.php” />
rel=”canoncial” tells search-engines the preferred version of a page. It’s only a suggestion, not a directive. As long as you’re split testing instead of doing a dodgy SEO tactic, Google is happy with canonical links and will listen. Be wary of implementing canonical site-wide. Bing seems to dislike site-wide canonical links saying, “Its best to leave them blank rather than point them at themselves.”
If you’re using WordPress, I recommend you install the Yoast plugin. Through its “Advanced” feature on a per-post and per-page basis, set the URL of the control page as the “Canonical URL”.
For old-schoolers without a CMS who edit separate files for each page, turn the handle of your power generator, boot up your Commodore 64, load up your FTP program, then manually add canonical code into your
section of the original and tested pages.4. Disappearing test assets
You found a winning version! Congratulations. You update your page with the best version then erase digital signs of your failed versions. Stop.
Firstly, by blindly deleting an old page you lose track of what you tested. The data you get from testing is the most powerful information you can obtain for a business. There’s no hypothesizing about what works. You know exactly what gets the business more sales. Never delete test information. Good split testing software should archive all test data and assets.
From an SEO perspective, when you delete an image of my face that was outperformed by a cute model, you create a void for Google. If Google tries to rescan an asset you deleted, your servers produces a 404 error telling Google the page, image, or video does not exist.
That product image you tested may show up number one in Google’s image results to bring you 100 visitors per month. Remove the image and Google will remove it from the search results causing you to miss the 100 uniques (and probably decent sales because the visitor is searching for the product).
I like to keep most test assets (videos, images, and audio) alive for these reasons. The only exceptions are unique pages setup solely for a test, which I delete then 301 redirect back to the control. We don’t want Google paying attention to these test pages remember. The deletion is more from my OCD for a clean system rather than a strategic purpose. As long as you canonicalize and use noindex, you’re fine.
Make it a monthly habit of checking your Google Webmaster Tools account for crawl errors.
5. Running the test too long
A test is a success when it achieves statistical significance – a term used by researchers and mathematical nerds to tell us “the observed data has a high-probability it would occur again”. No worries if you flunked out of Math in school or lost your Math skills like myself. Any decent testing tool should tell you if it is statistically significant.
It is a SEO best practice not to keep a split test running once you hit statistical significance. Google says:
If we discover a site running an experiment for an unnecessarily long time, we may interpret this as an attempt to deceive search engines and take action accordingly. This is especially true if you’re serving one content variant to a large percentage of your users.
Am I saying you better end a test to be safe the Big G does not hurt your rankings? Never, ever, ever, ever end a test before statistical significance in fear of SEO pain. Split testing noobs prematurely terminate tests.
Allow split testing to work its magic through these five best practices. Your bottom line depends on it.
For more help with improving conversions, I’ve written a complete guide to improve your ecommerce conversion rate. There you’ll learn more about how to split-test and optimize your website. I look forward to hearing your comments or questions about SEO and split testing below!
