A/B testing for SEO can't use Google Experiments or other methods that change pages on a per-user basis. You need to make sure that Googlebot and users each see a set of pages from group A and another set from group B. I generally pick half the pages on my site to change, and half to leave "as is".
This only works if you have a large number of similar pages on your site and are able to split them in half randomly. For example, I've done SEO experiment based on the database ID of the landing page. If the ID is divisible by 2 it goes into group A, otherwise it goes into group B.
This works particularly well with page title experiments and other on page factors. Group A gets one template for the title, Group B gets another.
Even then you may need to be careful to exclude your top pages when analyzing your results. For example if you have one page that goes viral when the experiment is running, a single page may be enough to tip the balance in favor of whichever group it is in.