SEO A/B testing (also called SEO split testing) checks whether a change affects organic search performance. The term has two meanings in practice. A conversion A/B test shows different versions of one page to different visitors. An SEO test splits a group of similar pages, for example product pages, into a test group that gets the change and a control group that does not, then compares organic clicks over time. Search engines cannot be shown a “version B” per visitor, which is why SEO tests work on page groups. Google does not describe this method itself; it is a general approach from practice.
How does an SEO test work?
- Pick a set of pages with a similar template and enough organic traffic.
- Split the set at random into a control group and a test group.
- Apply the change to the test group only, for instance a new title pattern.
- Compare the clicks of both groups against their pre-test trend.
The result is an estimate. Seasonality, updates and unrelated changes affect both groups, so only the difference between them carries information.
What does Google say about tests with variants?
Google’s guidance applies when visitors are sent to different versions of a page:
- No cloaking. Googlebot and users must see the same set of URLs. Showing different content to Googlebot is against Google’s spam policies.
- rel=“canonical”. Variant URLs should point to the original with rel=“canonical”. Google prefers this over noindex because it matches the intent.
- 302, not 301. Redirects to variants should be temporary (302) so the original URL stays in the index.
- Limited duration. Run the test only as long as needed, then remove variant URLs and testing scripts as soon as possible.
Example
An online shop has 400 category pages. It places 200 of them in a test group and adds the product count to the title. After four weeks, the clicks of the test group are compared with the control group. This setup is a made-up example.
Common mistakes
According to practitioner reports, these mistakes are common.
- Treating a CRO tool like an SEO test. Tools that swap content for visitors do not measure search effects.
- Test and control groups that differ. Pages with different traffic patterns make the comparison unreliable.
- Leaving test code live. Variants and scripts that stay after the test can look deceptive to Google.
Sources
- Website testing and Google Search – Google Search Centraldevelopers.google.com
Last updated: