How much traffic does an A/B test actually need?

More than most enterprise B2B pages have. The figure follows from two inputs: your current conversion rate, and the size of the improvement you want to be able to detect. The smaller the improvement, the more traffic it takes to see it at all. So we start from your numbers rather than a rule of thumb. A public significance calculator will give you the arithmetic in a minute, and it is usually sobering. If a page cannot get there, you should hear it before you buy a testing licence, not after. Those inputs come from your analytics, which is why measurement gets verified before any experiment is designed.

What can we actually test on a WordPress page?

On a site running Gutenberg: individual blocks, whole sections, complete landing pages, headlines, images, calls to action, forms, menus and widgets. Block-level tests are the cheapest to run and the easiest to attribute. Page-level tests answer bigger questions and need proportionally more traffic. Menu and navigation tests touch every page, so they need a longer window and closer monitoring. One variable per test, always. Change two things at once and a win tells you nothing about which change earned it.

Will running experiments slow the site down?

A little, and it is worth being deliberate about it. Every testing tool loads tracking script on the frontend, and that cost lands on every visitor while the benefit only lands on the winning variant. Two things to watch. Script weight on the critical rendering path, which shows up directly in Core Web Vitals. And full-page caching serving a cached variant to a visitor who was assigned the other one, which does not break the page, it corrupts the result. Both are handled during setup, alongside the rest of the work on our performance page.

Does A/B testing create a GDPR problem?

It creates the same obligation as any other tracking. Experiments assign visitors to variants and record what they do, so where GDPR applies the tool sits behind your cookie consent banner rather than beside it. The practical consequence is that you only measure consented traffic, which shrinks your effective sample and lengthens the test. That is a reason to allow for it when sizing the experiment, not a reason to cut the consent step.

Does A/B testing hurt SEO?

Not if the variants are handled properly. Google's guidance on this has been stable for years: Point a canonical at the URL you want indexed. Use a temporary redirect, not a permanent one, for redirect tests. Serve crawlers the same thing you serve visitors. Take the test down once it is finished. The real failure mode is operational rather than technical. A test left running for a year, until nobody remembers setting it up, has quietly become permanent cloaking.

Who runs the experiments after handover?

Your marketing team. An editor who can build a landing page can build a variant of it, and simple variants need no developer involvement at all. What we hand over alongside the configuration is the discipline: a record of which tests are running and which have expired, a written result for every experiment including the ones that decided nothing, and the sizing judgment that says which page is worth testing next.

A/B Testing

Enterprise WordPress A/B testing, sized to the traffic you actually have

The hard part is not the tool. It is that most enterprise B2B pages never collect enough conversions to separate a real winner from noise. We start by working out which of your pages can carry a test.

Home
Capabilities
A/B Testing

What has to be in place

An experiment is a measurement problem before it is a design problem

Five conditions decide whether a test returns an answer or just a number. Getting them wrong does not produce a failed test. It produces a confident wrong one.

Enough traffic to reach significance

Sample size follows from your current conversion rate and the lift you want to detect. We count sessions and conversions per page first: on a low-traffic page a test never resolves, and the decision gets made on impatience instead.

One variable, one conversion goal

A headline or a call to action or an image, not all three at once, measured against one defined event: a click, a form submission, a page view. Everything else is recorded, and none of it gets to pick the winner afterwards.

Variants built in Gutenberg, not in a ticket

Nelio works at block level, so a marketer can duplicate a FAS block, change the copy and publish the variant. Simple variants need no development involvement, which is the reason for keeping the tooling in the CMS.

Tracking accounted for in performance and consent

Every testing tool adds script weight to the frontend, and where GDPR applies the tracking sits behind your cookie banner like any other tag. Both get decided during setup, not after the first experiment goes live.

Hosting that allows the tracking to run

Testing tools execute script in the visitor’s browser and report to an external service, and not every managed hosting setup permits that by default. We confirm your environment supports it before a licence is bought, not during the first experiment.

How it works

Four phases, and the first one can end the project

If the traffic is not there, the honest recommendation is to spend the effort somewhere else. We would rather say that in week one.

1

Check volume and measurement

We count traffic and conversions per candidate page, and confirm the conversion event fires and survives the consent banner. Pages that cannot reach significance in a sensible window get named as such.

2

Pick the tool and define the goal

Nelio by default, or an external platform where the scope justifies it. The conversion objective is configured before anything is built: a CTA click, a form submission, a specific page view.

3

Build the variant, one variable at a time

Directly in Gutenberg or from the plugin panel, changing a single element so the result is attributable. Blocks, sections, whole landing pages, titles and menus are all testable.

4

Run it, then close it

Seven to fourteen days depending on traffic. The winner replaces the original, the test is deactivated with its history kept, and the result gets written up, including the experiments that decided nothing.

Choosing the tool

Three options, and the reason the choice is open again

Google Optimize was discontinued in 2023, which is why so many enterprise teams are picking an experimentation platform from scratch rather than inheriting one.

Nelio A/B Testing

Our default. A premium WordPress plugin that tests pages, posts, individual Gutenberg blocks, titles and menus, with device-level segmentation and a results panel in the admin. Priced by page count and traffic volume, and limited for genuinely multivariate work.

Optimizely or VWO

External SaaS platforms. More powerful, multichannel, and correspondingly more expensive and more complex to operate. Worth it when experimentation spans more than the website and somebody owns it as their job.

Custom JavaScript with Tag Manager

Maximum flexibility and the highest ongoing effort. It is not editorial-friendly: every variant becomes a developer task, which undoes most of the reason for testing inside WordPress in the first place.

What enterprise teams ask us about A/B testing

Worth reading even if you never hire us.

Related insights

Keep reading on conversion and landing pages

Landing page practice, optimization workflows, and what to change when a test is not an option.

Manager planning landing page campaign at table

Apr 7, 2026

WordPress

Request an experimentation readiness review

We will check your conversion tracking, count the traffic on your key pages, and tell you which can carry an experiment and which cannot. You get the findings whether or not you work with us.

Request a readiness review Talk to a WordPress conversion specialist