A/B testing in 2026: what to use instead of Google Optimize

For years Google Optimize was the default answer to “what do I start A/B testing with” — it was free and a GA account was enough. Google switched it off in September 2023, and the web is still catching up: plenty of guides you’ll find in search still recommend a tool that hasn’t existed for over two years.

This article is the current answer: what to use instead, how to choose by site size and budget — and what to do when your traffic makes “just run an A/B test” easy to say and hard to measure.

A quick reminder: what an A/B test is and isn’t

An A/B test splits visitors randomly into two groups: one sees the current page (the control), the other a modified variant. After a predetermined period you compare which group completed the conversion more often — and because the split was random, the difference can be attributed to your change rather than to chance or season.

What an A/B test is not: looking at analytics “before and after” you shipped a change. That comparison also measures everything else that changed in the meantime — campaigns, seasonality, the weather. Which is exactly why you test in parallel, not in sequence.

What to replace Google Optimize with

After Optimize ended, the market settled into three categories by budget and technical setup.

1. Paid visual tools — closest to the spirit of Optimize

VWO, AB Tasty and Convert are the most direct replacements: a visual editor (you change the page by clicking, without a developer), WYSIWYG previews, GA4 integration. They’re billed monthly by traffic; for smaller sites the limiting factor tends to be exactly that price, which only makes sense above a certain volume of traffic and conversion value.

Optimizely is the enterprise tier — powerful, but beyond the needs of most sites that used to run on Optimize, in both price and complexity.

2. Product platforms with experiments included

PostHog is probably the most interesting choice today for more technical teams: experiments are part of a platform that also covers analytics, session recordings and feature flags, and the free tier is generous enough that a smaller site pays little or nothing. The downside: no visual editor — variants are defined through feature flags, so you need a developer, or at least a snippet and some nerve.

The open-source GrowthBook goes in a similar direction — free on your own infrastructure, connected directly to your data (GA4/BigQuery, a warehouse). Ideal when you’d rather not send data to a third party.

3. Testing built into what you already use

E-commerce platforms and landing-page tools (Webflow/Framer optimisation, email platform builders, store add-ons) are gradually adding simple built-in tests. If you need to test one landing page and nothing more, the built-in option is often the fastest route — just expect simplified statistics and results that travel less well.

How to choose

• Have a developer (or willing to touch code)? PostHog or GrowthBook — the best value and full control.
• Need a marketer to build tests without a developer? VWO or AB Tasty — you’re paying for the visual editor.
• Testing a single page once? Whatever is built into your platform, if anything.

But the tool is the smaller half of the decision. What you test, and how, matters more.

How to test on low traffic

Most guides go quiet here, even though it’s the reality for most sites: hundreds to low thousands of visits a month on the page being tested. Honest maths says a small sample only reveals large differences — and four practical rules follow from that.

1. Test big changes, not details. A different headline and offer in the hero, a restructured page, a shorter form — changes where a double-digit effect is plausible. Button colour moves conversion by a couple of per cent, and your sample won’t be enough for that.

2. Test where the most people flow. The homepage, the main landing page, the entry to checkout. A test on a page with a fifth of the traffic runs five times longer.

3. Let the test finish. Calculate the required sample in advance (every decent tool has a calculator) and until then don’t treat interim numbers as results. Watching “it’s up 30 %!” after three days is the most reliable way to walk away with a false conclusion.

4. When the sample is hopelessly small, don’t test — measure differently. Below a certain traffic level it’s more honest to make a change grounded in analysis of where visitors drop off, ship it, and watch a longer trend. It isn’t an A/B test and it doesn’t carry the same proof — but it beats a test that never reaches significance.

The most common mistake: testing without a hypothesis

Tools tempt you into “let’s test it and see”. But a test without a hypothesis can’t be interpreted: even when the variant wins, you don’t know why, and next time you start from zero again.

A good hypothesis has a shape: because we see X in the data, we believe change Y will increase Z. For example: because most visitors leave the form page within 30 seconds, we believe cutting the form from nine fields to four will increase submissions. The test then confirms or refutes it — and either way you learned something.

Where do hypotheses come from? From data and from watching customers: funnel analysis shows where people get lost, user testing hints at why.

---

The takeaway

The end of Google Optimize turned out to be a useful filter: it forced people to pick a tool by actual needs rather than by a price of zero. Technical teams are best served by PostHog or GrowthBook, marketing teams without a developer by VWO or AB Tasty — and for everyone, the tool isn’t what decides. What decides is a hypothesis grounded in data, a large enough sample, and the discipline to let the test finish — how much traffic that takes is in the calculator in why your A/B test lies.

And if the work of running all that is what’s holding you back: that loop — data → hypothesis → test → result — is what Kaen runs for you continuously. It finds where the funnel leaks, proposes tests with a hypothesis attached, and ships them once you approve.