10
BEST PRACTICES: OPTIMIZING A/B TESTS FOR SEO BY LAURENT MÜLLENDER, 2015

Best practices: Optimizing A/B tests for SEO

Embed Size (px)

Citation preview

Page 1: Best practices: Optimizing A/B tests for SEO

BEST PRACTICES:OPTIMIZING A/B TESTS FOR SEO

BY LAURENT MÜLLENDER, 2015

Page 2: Best practices: Optimizing A/B tests for SEO

A/B TESTING AND SEO

Although Google officially states that A/B Tests should not impact SEO performance, there is still a risk of:

Duplicate contentIndexed variantsDe-listing/penalty

On the following slides I will show you:

How to avoid or minimize those risks.How to avoid common mistakes.How to properly setup A/B (or multivariate) tests so that there is no negative impact on your SEO performance.

Page 3: Best practices: Optimizing A/B tests for SEO

AVOID DUPLICATE CONTENT

When testing, you usually test two or more pages with similar content but slight changes. This could be seen as duplicate content by search engines.

Use the same URL for the test. Either through DOM manipulation or parameters.

Control: www.domain.com/test.html

Variant: www.domain.com/test.html

Or

Control: www.domain.com/test.html?v=1

Variant: www.domain.com/test.html?v=2

Avoid using two different pages, under two different URLs

Control: www.domain.com/control.html

Variant: www.domain.com/variant.html

Page 4: Best practices: Optimizing A/B tests for SEO

AVOID INDEXING OF TESTS (1/2)

If you use different URLs for the test, there‘s a risk that the test page(s) gets indexed.

To avoid this, you can set a canonical to the original page and set the test page to NOINDEX. Note: Google prefers canonical over NOINDEX because it more closely matches your intent in this situation.

Example:

On variant.html:

<link rel="canonical" href="https://www.domain.com/control.html" />

and

<META NAME="ROBOTS" CONTENT="NOINDEX">

Important: should you use redirects to split traffic, use 302 instead of 301 redirects. This tells Google that this redirect is just temporary.

Page 5: Best practices: Optimizing A/B tests for SEO

AVOID INDEXING OF TESTS (2/2)

You can also avoid indexing of the test page(s) by excluding Search Engines from your test so that they always see the original version. Additional benefit: bots might skew your test results.

1. In your ROBOTS.TXT file:

User-Agent: Googlebot

Disallow: /variant.html

2. In your testing tool:

In most testing tools you can exclude certain types of traffic from the test. You can exclude bots like Googlebot with a regular expression. Example Visual Website Optimizer:

Please keep in mind that Google wants to visit a page just like a regular user. In times of marketing automation, personalization and massive testing, Google has softened its rules, but it could still be seen as content cloaking.

Page 6: Best practices: Optimizing A/B tests for SEO

WHAT IF A VARIANT GOT INDEXED?

You can check the indexing status of a page using the following search:

Should your page appear, you can either wait for the next crawling, or remove the URL from the index using Google‘s Search console (formerly known as „Webmaster tools“)

Page 7: Best practices: Optimizing A/B tests for SEO

SHORTEN THE TESTING TIME

Do not let your test run for too long. Here‘s Google‘s view on long tests:

“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.” (Google Webmaster Central blog)

When preparing your test you will calculate the optimal sample size for a given significance level that will be validated with a t-Test. Knowing your traffic and seasonal effects, you can then calculate the time needed to reach the calculated sample size. A calculation is always better than gut feeling and you keep the length to the strict minimum.

However, t-Tests are slow and are only useful with an end point to collecting data, a fixed sample. Sequential testing algorithms are more suited for testing as they analyze and evaluate data as they come in. Bayesian models return ranges that become more and more accurate over time and they can be looked at at an early stage of the testing time.

Such models help you take a decision faster than with t-Tests and thus it reduces testing time. Modern testing tools such as Visual Website Optimizer or Optimizely support such algorithms.

Page 8: Best practices: Optimizing A/B tests for SEO

AFTER THE TEST

Do not just delete the variant.

This might generate a 404 error code should the page or elements of the page such as images have been indexed or linked.

301 redirect the variant URL to the winning URL.

Page 9: Best practices: Optimizing A/B tests for SEO

IN A NUTSHELL

1. Use a single URL for your test to avoid duplicate content risks.

2. If you use different URLs for you split test, set a canonical and NOINDEX for your variants.

3. If you use redirects to split traffic, use 302 instead of 301.

4. In some cases you might want to exclude bots from your test.

5. Should a page get indexed, remove it from the index through the Google Search console.

6. Shorten testing time by calculating the optimal testing length or using better statistical methods.

7. After the test, 301 redirect the URLs to the winning URL to avoid 404 errors.

“I've been quite clear that there's nothing wrong with A/B testing. In fact, less than a month ago I tweeted that "A/B testing can be really helpful" and included a link to best practices: https://twitter.com/#!/mattcutts/statuses/191658511149711360 “(Matt Cutts, Source)

Page 10: Best practices: Optimizing A/B tests for SEO

AUTHORLaurent MüllenderPerformance-driven product management and online [email protected]