WordPress plugin
Most WordPress A/B testing breaks the moment you turn on page caching: everyone gets variant A, or the original flashes up and swaps half a second later. This one decides in the browser, against a cached page, before anything is painted.
The variant is chosen in the browser against one cached copy of the page, with the same cache-plugin exclusions we ship for WP Rocket, WP Fastest Cache, LiteSpeed, Autoptimize and Cloudflare Rocket Loader. You do not have to choose between a fast site and a tested one.
Only the elements a variant actually changes are hidden, for the moment it takes to change them — so visitors who get the control wait for nothing at all. A failsafe reveals everything regardless after 1.2 seconds, which means a mistyped selector can never leave part of your page blank.
Below 300 views and 25 conversions a variant, it reports no comparison and tells you how many views you would need. Run more than one challenger and the bar rises automatically. Under a week, it says the test is short.
Clicks anywhere, plus WooCommerce add-to-cart and completed orders with revenue — counted from the order itself, server-side, where they cannot be faked from a browser console. Refunds come back off again.
Putting someone in a test means storing an identifier, which under UK GDPR and PECR is not strictly necessary. By default nobody is bucketed until statistics consent is given, via the WP Consent API or our own banner. You can change that; the settings screen explains what you take on.
A test is a CSS selector, what to change it to, and what counts as success. No page builder, no visual editor to fight, and no dashboard on somebody else's server holding your results.
The awkward part
Every other split testing tool shows you a percentage from the first day. On a hundred visitors that percentage is noise, and acting on it is worse than not testing at all — you make a change, you believe it worked, and you carry the belief into the next decision.
This plugin reports nothing until there is enough data to mean something, and says how many views that would take at your conversion rate. When there is enough, it gives you the difference, a confidence interval and a p-value — and if more than one variant is running, it raises the bar, because testing several at once makes a false winner more likely.
Most of the time the honest answer is “no difference worth acting on”. That is a result: it means the change is not worth making, and you can stop wondering about it.
FAQ
The engine is a couple of kilobytes printed inline in the head. There is no external request before a variant is chosen, because waiting for one is what causes the flash this plugin exists to prevent.
Nothing visible. Elements are hidden only while a change is applied, and a failsafe reveals them after 1.2 seconds whatever happens. A bad selector means the test does nothing — it can never leave a blank space where your headline was.
Because early differences are almost always noise. On 120 views a variant can look 58% ahead and be worth nothing, and a tool that shows you that number has taught you something false. It will tell you what it would take to know, then tell you when you know it.
Yes. Add-to-cart and completed orders are recorded from the order itself rather than the browser, revenue is attributed per variant, and refunds are subtracted again — a variant that wins on orders which come back has not won anything.
In your own WordPress database, as daily counters per variant. There is no per-visitor event log, nothing is sent to us, and no row could be traced back to a person.
Automatic updates and support, on up to 5 websites. Your tests keep running whether or not the licence is active — a lapsed key pauses updates, it never takes a live test off your site.
Up to 5 websites per licence. 14-day money-back guarantee.