GoHighLevel split testing, without fooling yourself

Split testing is included in GoHighLevel on every plan, it lives inside the funnel builder, and it takes about four minutes to configure. The hard part is not the setup. The hard part is not talking yourself into a winner that was noise.

This is how to run one that produces a decision you can act on, and how to recognise the tests that were never going to.

What the feature actually does

A funnel step can hold more than one page. When it does, GoHighLevel divides incoming traffic between those pages by a percentage you set, and records views, opt-ins and sales per variant. Visitors are held on their assigned variant so the same person does not see both.

One constraint decides whether this is available to you at all: split testing is a funnel feature. Website pages cannot be split tested. If the page you want to test is on the sites side, it has to move — which is one of the sharper edges in GoHighLevel funnel vs website.

Setting one up

1. Open the funnel and the step you want to test. Duplicate the existing page — that duplicate is variant B and the original is your control.
2. Change exactly one thing on variant B. One. This is the rule the whole method rests on.
3. Set the traffic split. Fifty-fifty unless you have a reason; an uneven split takes longer to reach the same confidence.
4. Decide the success metric before you start it, and write it down. Opt-ins, bookings or sales — not page views, and not "it feels better".
5. Publish both variants and confirm both actually load on the live URL. A variant that errors will quietly take its share of traffic and convert nobody.
6. Leave it alone.

Step six is the one people fail. Checking a test on day two and switching the split because one variant is ahead does not accelerate the result; it destroys it.

What to test, in order of value

Test in descending order of how early the element appears and how much depends on it. Everything below the headline depends on the visitor still being there.

Most people start at row six because it is the easiest change to make. On a page with modest traffic a button colour test will never reach a conclusion, and the months spent waiting were months not spent testing the headline.

If you do not know which block is losing people, the block order and job of each one is in high converting landing page anatomy — reading that first often removes the need for the test entirely, because the fault is a missing block rather than a weak one.

How long to run it

Long enough for two things: enough conversions to be more than luck, and enough calendar time to cover your traffic's weekly rhythm. A minimum of one full week, always, because a test that starts on Wednesday and ends on Friday compares midweek traffic to nothing.

On volume, the useful discipline is to think in conversions rather than visitors. If a variant has produced four bookings and the other has produced seven, that is not a fifty-per-cent improvement — it is three bookings, and three is a number that moves on its own. Wait until both variants have accumulated enough conversions that a handful either way would not change your conclusion.

We deliberately do not publish a required sample size here, because the honest answer depends on your baseline rate and the size of the difference you care about, and any single number would be wrong for most readers. Use a sample-size calculator with your own baseline rate as the input. Your baseline comes from your own funnel — set your own conversion baseline rather than borrowing an industry figure, because a wrong baseline produces a confidently wrong sample size.

The five ways a split test lies to you

• Two changes at once. If you change the headline and the image, a win tells you nothing about which one to keep — and the two may be cancelling each other out.
• Stopping the moment one variant is ahead. Early leads reverse constantly. The result only means something at the point you decided in advance to stop.
• Changing the traffic split mid-test. The two variants are now measured over different periods with different traffic. The comparison is void.
• Running through an anomaly. A holiday, an outage, a viral post or a competitor's sale all break the assumption that the only difference between variants is the variant.
• Measuring the wrong step. A headline that lifts opt-ins and reduces bookings is a losing headline. Measure at the step that produces revenue, not at the step you changed.

That last one deserves emphasis, because it is the failure that produces the most confident wrong decisions. Track the whole path, which means the tracking has to reach past the page you are testing — see GoHighLevel conversion tracking.

When not to split test at all

Below roughly a few hundred visitors a month, a split test will not reach a conclusion before the market has moved. That is not a reason to stop improving the page — it is a reason to improve it by changing it decisively, one substantial change at a time, and comparing month against month while acknowledging that the comparison is not clean.

It is also the point at which the honest advice is to spend the effort on traffic and on offer instead. A page that converts poorly with fifty visitors has a fifty-visitor problem, and no amount of testing will fix it.

Recording what you learn

Keep a plain list: date started, date ended, what changed, the metric, the numbers, and the decision. Three tests in, this is more valuable than any of the individual results, because it stops you re-running a test you already ran and it shows you which kinds of change move your audience.

Then roll the winner into the control and start again. A funnel that has been through six honest tests is a meaningfully different asset from one that has been through none, and the compounding is the entire point.

If you are testing a page inside a snapshot, duplicate the step rather than editing the original, so you keep a clean copy of the build you started from. Every system in the catalogue is structured to make that duplication safe — the webinar funnel snapshot in particular, where the registration page is the step almost everybody tests first.