Splitlab home

Book a Demo

Splitlab home

Book a Demo

Splitlab home
The splitlab.io features page in a desktop browser, between the Splitlab Traffic Split menu (Equal Distribution, Custom Distribution, Multi Arm Bandit) and the Metrics list with Revenue per Visitor at the top

November 18, 2026

By

Bishal Katuwal

·

15 min

read

Landing Page Split Testing: The Complete Guide for Funnel Pages

November 18, 2026

By

Bishal Katuwal

·

15 min

read

Landing Page Split Testing: The Complete Guide for Funnel Pages

In the ever-evolving business environment, achieving a competitive edge demands innovative solutions beyond conventional methods. Organizations require smart systems that can optimize workflows, enhance customer interactions, and reveal new avenues for growth. Enter Biotix, the cutting-edge AI-powered CRM management tool.

The three kinds of split test and when each one fits, how to split the traffic (equal, custom or bandit), what to test first on a funnel page, how much traffic you really need, and the mistakes that fake a winner.

Bishal Katuwal

Bishal Katuwal

Splitlab Founder

The splitlab.io features page in a desktop browser, between the Splitlab Traffic Split menu (Equal Distribution, Custom Distribution, Multi Arm Bandit) and the Metrics list with Revenue per Visitor at the top
The splitlab.io features page in a desktop browser, between the Splitlab Traffic Split menu (Equal Distribution, Custom Distribution, Multi Arm Bandit) and the Metrics list with Revenue per Visitor at the top

Landing page split testing means sending your visitors at random to two or more versions of a page and keeping the one that makes more money. That’s the whole idea. The hard parts are choosing the right kind of test, testing things big enough to matter, and not fooling yourself when you read the result.

And remember who you’re testing for. In my paid-ad funnels, 90% or more of visitors are on a phone. Build every version for the phone first, and check it on a phone before it goes live.

This is the reference guide: the types of test, what to test first on a funnel page, how much traffic you need, and the mistakes I see most. If you want the step-by-step setup instead, read how to A/B test landing pages.

Key takeaways

  • There are three kinds of split test. A/B on the same URL, split URL and multivariate. Pick by what you’re changing and how much traffic you have.

  • A multi-armed bandit isn’t a fourth kind of test. It’s a way of splitting the traffic, and you use it together with an A/B or multivariate test.

  • Test the offer before the button. Price, promise, proof and guarantee move revenue. Button colors rarely move anything you can measure on a small page.

  • Traffic decides what you can test. Halve the lift you want to detect and you need four times the visitors.

  • Most fake winners come from a handful of mistakes. Stopping early, changing too much, the wrong metric and revenue you can’t track.

  • Judge every test on revenue per visitor. A version that gets more buyers can still make less money.

What is landing page split testing?

A split test shows different versions of a page to different visitors at the same time, chosen at random, and compares what each group does. Because the visitors are split at random and see the pages over the same days, the only real difference between the groups is the page. So if one group buys more, the page is the likely reason.

That’s what makes it better than “before and after”. If you change your sales page on Monday and sales rise on Tuesday, it might be the page. It might also be a better ad, a podcast mention or payday.

People use “split test” and “A/B test” to mean the same thing. Strictly, a split URL test is one kind of A/B test. More on that below.

The three types of split test, and when to use each

There are two separate choices in every test. What you test (the type of test) and how you split the traffic (equal, custom or multi-armed bandit). People often mix them up, so I’ll take them one at a time.


Two separate choices in every split test. What you test: an A/B test on the same URL for one change, a split URL test for a whole new page, or a multivariate test for several elements in every combination. How you split the traffic: equal, custom, or multi-armed bandit, which shifts traffic to the version earning more. You pick one from each column, for example an A/B test with a bandit split

1. A/B test on the same URL

Visitors land on one URL and the tool shows them version A or version B of that page. Use it for changes to one part of the page: a headline, a price, a guarantee block, a shorter form.

Why it’s the default: it keeps one URL, so your ads, emails and tracking stay the same. It also gives the cleanest answer, because only one thing differs.

Example: your $27 ebook page says “Write your first book in 30 days”. Version B says “Finish your first draft in 30 days, even if you’ve never written”. Same page, same URL, one change.

2. Split URL test

Each version lives at its own URL, like /webinar and /webinar-v2, and the tool sends each visitor to one of them. Use it for a whole new page: a new layout, a different funnel builder template, a long page against a short one.

Why: a full redesign is easier to build as a separate page than as edits layered on the old one. In ClickFunnels or HighLevel, that usually means duplicating the step and editing the copy. The trade-off is that you learn which page won, not which change made it win.

3. Multivariate test

You test several elements at once, and the tool builds every combination. Three headlines, two hero images and two guarantees make 3 × 2 × 2 = 12 versions. Use it when you have a lot of traffic and want to know which elements matter and how they work together.

Why it’s powerful: it finds the best combination, not just the best headline. A headline that wins alone might not work with the image that wins alone. The catch is traffic: it’s split 12 ways instead of 2, so each version takes longer to reach a readable result. Pair it with a multi-armed bandit split and the weak combinations get less traffic early, which makes it workable on smaller pages. I explain the trade-off in A/B testing vs multivariate testing.

How to split the traffic

Once you know what you’re testing, you pick how the visitors are divided. In Splitlab you choose this per test.

  • Equal split. Every version gets the same share. The cleanest read, and the right choice when you want to learn a rule you’ll use on other pages.

  • Custom split. You set the shares yourself, like 80/20. Use it to limit the risk when a new version is a big change to a page that already sells.

  • Multi-armed bandit. It starts even, then sends more visitors to whichever version is earning more while the test runs. Use it for short promotions, paid traffic and many versions at once.

Think of it like a motorbike and its tires. You can’t ride without tires, but you can choose thin ones, thick ones or off-road ones. The A/B or multivariate test is the bike. The split is the tires. So the question is never “bandit or A/B test?”. It’s “A/B test with an equal split, or A/B test with a bandit split?”


The Splitlab Traffic Split menu with three options: Equal Distribution for an even split and a clean read, Custom Distribution where you set each share, and Multi Arm Bandit, which shifts traffic to the higher-revenue variant

I compare all three in equal vs custom vs multi-armed bandit, and go deeper on the bandit in multi-armed bandit vs A/B testing.

What to test first on a funnel page

On info-product and course pages, I rank test ideas by how much they can change what a visitor is willing to pay. Here’s the order I work through, top first.


What to test first on funnel and info-product pages, biggest impact on top: 1 offer and price ($997 vs 3 × $347), 2 headline and promise (repeat the ad), 3 proof (a result by the button), 4 form length (email only?), 5 page length (long vs short), 6 guarantee (a specific refund). Button colors are crossed out: skip on low traffic, too small to measure

1. The offer and the price

Why first: the offer is what people buy. Everything else on the page just explains it. A different price, a payment plan or a bonus can change revenue per visitor more than any copy change.

I learned this on my own order pages. One test changed nothing but the offer around the checkout: a pre-checked order bump that showed its discount, and a line of reviews under the button. Same traffic, same page. Earnings per click went from $3.03 to $4.68, and average order value from $35 to $39.

Why This Tiny Order Page Change Almost Doubled My EPC

Example: test a $997 course against three payments of $347. Or test the same $27 ebook with and without a $17 workbook bump on the order form. Fewer people may buy, but each buyer may spend more, which is why you judge it on revenue.

2. The headline and the promise

Why: the headline decides whether people read anything else. On a phone, the first screen is often just the headline, one line and a button. If the promise doesn’t land there, the rest of the page doesn’t get a chance.

Example: an Instagram ad says “Get your first coaching client in 14 days”. If the page headline says “The Coaching Accelerator Program”, the visitor has to work out whether they’re in the right place. Test a headline that repeats the ad’s promise. My guide on how to test landing page headlines has more on writing the versions.

Small wording changes can move real money. Harvard Business Review reported that a change to how Bing displayed ad headlines increased revenue by 12%. Bing has search-engine traffic, though. Your page probably needs a bigger change to show a result you can read.

3. Proof

Why: once people want the result, the next question is “will it work for me?”. Proof answers it. Where the proof sits matters as much as what it says.

Example: move your best student result, with a name, a photo and a number, from the bottom of the page to just above the first buy button. Or test a webinar replay clip against written testimonials.

4. Form length

Why: every field is a small reason to stop. On an opt-in or application page, the form is the last thing between interest and a lead.

Example: test a webinar registration with name and email against email only. Or test a coaching application with 12 questions against a two-step version with 4 easy questions first. Judge it on booked calls or sales later, not just sign-ups, because shorter forms can bring in less qualified people.

5. Page length

Why: the right length depends on the size of the decision. A free checklist can sell itself in one screen. A $997 course usually needs room for objections, curriculum and proof.

Example: test a long-form sales page against a short page with a video and a buy button. Run this as a split URL test, because it’s a different page, not a different headline.

6. The guarantee

Why: a guarantee lowers the risk of saying yes, and people weigh risk most right at the button. It’s also cheap to test, because it’s one block of copy. But don’t assume it wins. When I added a 30-day guarantee to a $5 book funnel, checkouts dropped from 47% to 31%, in two separate funnels.

Example: test “30-day money-back guarantee” against a stronger, more specific version, such as “Do the first two modules. If you don’t have a finished outline, email us for a full refund.” Watch refunds alongside sales, because a stronger guarantee can raise both.

What not to bother testing on low traffic

Button colors, icon styles, font tweaks and one-word changes to a button label. On a big site these can be worth testing. On a funnel page with a few thousand visitors a month, the difference they make is usually too small to measure, so the test runs for months and ends in a tie.

Why that matters: every week a test runs is a week you didn’t test the price or the promise. Make the button easy to see and move on. I cover what color can and can’t do in best colors for conversion.

How much traffic you need, in plain words

It depends on your traffic. The size of the change you want to detect decides how many visitors you need, and your daily traffic decides how long that takes. Big differences show up quickly. Small differences need a lot of traffic. Spotify’s experimentation guide puts the rule simply: “halving the MDE quadruples the required sample”. The MDE is the smallest lift you want the test to be able to detect.


Visitors needed per version on a page converting at 5%: about 8,200 to detect a lift to 6%, about 31,000 to detect 5.5%, and about 122,000 to detect 5.25%, the size of a tweak like a button color. Halve the lift you want to detect and you need four times the visitors. Two-proportion formula at 95% confidence and 80% power; 5% is about a $27 page in Bishal Katuwal’s funnels

Here’s the math with the standard two-proportion formula (95% confidence, 80% power), the same one I use in the step-by-step guide, for a sales page that converts at 5%, which is about where a $27 product sits in my funnels:

  • To spot a lift from 5% to 6%: about 8,200 visitors per version, so around 16,400 in total.

  • To spot 5% to 5.5%: about 31,000 per version.

  • To spot 5% to 5.25%: about 122,000 per version. That’s the size of lift a button color might give you, if it gives you anything.

Opt-in pages are kinder. A simple name-and-email page should convert at 20% or more, and at 20%, spotting a lift to 24% takes about 1,700 visitors per version. That’s why I test the opt-in page or the offer before small changes on a low-converting sales page.

In plain words: if your page gets 1,000 visitors a week, only test changes big enough to move conversion by a fifth or more. Anything smaller won’t finish. With high traffic, a week can be enough. With low traffic, plan on a month.

Common split testing mistakes, and why they fake a winner


Before you trust a winner, six checks: ran the full planned sample, not stopped on an early lead; one change per A/B test; judged on revenue per visitor, not clicks or opt-ins; revenue tracked through checkout with orders, bumps and upsells credited; no launch spike in the window; the same traffic mix on each version, one link split at random

Stopping the test early

Why it’s a problem: results swing a lot in the first days. If you check every day and stop the moment one version looks significant, you’ll often stop on luck. In Evan Miller’s example of a test with no real difference, checking after every visitor and stopping at the first “significant” result gave a false winner 26.1% of the time, “more than five times what you probably thought the significance level was.”

Fix: decide the sample size and the length (full weeks) before you start, and don’t act before then.

Changing too many things at once

Why: if version B has a new headline, a new price and a new video, and it wins, you don’t know which change did it. Next time you can’t repeat the good part and drop the rest.

Fix: change one thing in an A/B test. If you really want to change several, call it a split URL test of a new page and accept that you learn less, or run a multivariate test if you have the traffic.

Judging on the wrong metric

Why: the easiest number to move is the one closest to the page, like clicks or opt-ins. But a cheaper-sounding promise can get more opt-ins and fewer buyers. I walk through real ways this happens in why your A/B test winner didn’t make you more money.

Fix: pick the main metric before the test starts. For pages that sell, that’s revenue per visitor.

Not tracking revenue across the checkout

Why: a click on “Buy now” isn’t a sale. Baymard Institute’s research puts the average cart abandonment rate at 70.22%. And in a funnel, a lot of the money comes after the first purchase, in the order bump, upsell and downsell. If your tool only sees the landing page, it’s judging on the part of the funnel that makes the least money.

Fix: connect your checkout so every order, bump and upsell is credited to the version the buyer saw.

Testing during a launch spike

Why: launch traffic isn’t normal traffic. Your warm email list on cart-open day buys at a very different rate from cold Meta ad traffic in a normal week. A version that wins with your biggest fans may lose with strangers, and the spike ends before the test can tell you.

Fix: run tests on steady, evergreen traffic. During a launch, use a bandit to earn more, not to learn a rule.

Uneven traffic sources

Why: in the funnels I run, paid traffic converts lower than organic and webinar viewers. If one version gets more email traffic and the other gets more ad traffic, you’re comparing audiences, not pages. This happens when people send one link to their list and another to their ads.

Fix: send all traffic through one entry point and let the tool split it at random. Then check the result by source before you trust it.

How to read the results on revenue per visitor

Revenue per visitor (RPV) is total revenue from a version divided by the number of visitors who saw it. It combines how many people buy with how much each one spends, so it’s the number that actually pays your ad bill.


The Splitlab Metrics screen with Revenue per Visitor (RPV) selected as the main metric, Conversion Rate (CVR) and Average Order Value (AOV) alongside, and custom metrics for the order page and Upsell 1

Here’s an example of why it matters. The numbers are made up to show the math:

  • Version A: 1,000 visitors, 50 buy the $27 ebook (5%), 10 add the $97 upsell. Revenue $2,320. RPV $2.32.

  • Version B: 1,000 visitors, 60 buy the $27 ebook (6%), 6 add the $97 upsell. Revenue $2,202. RPV $2.20.

B wins on conversion rate and loses on money. Maybe its headline promised something easier, and the people it brought in were less ready for the upsell. If you only looked at conversion rate, you’d ship the version that earns less.

When you read a result, check three things:

  1. Did it reach the sample size you planned, over full weeks? If not, it’s not done.

  2. Which version leads on revenue per visitor, and how sure is the tool? Look at the probability of being best and the range, not just the average.

  3. Did a guardrail get worse? Refunds, chargebacks or no-shows on booked calls can wipe out a win.

I go deeper on the trade-off in revenue per visitor vs conversion rate.

How Splitlab helps

Splitlab runs A/B, multivariate and multi-armed bandit tests on funnel pages, including ClickFunnels 2.0, ClickFunnels Classic and GoHighLevel. Its integrations credit every order, bump, upsell and booked call to the version the visitor saw, and it reports revenue per visitor next to conversion rate. See the full feature list or book a demo.

FAQ

What is the difference between A/B testing and split testing?

Most people use them to mean the same thing. When there’s a difference, an A/B test changes part of a page on one URL, and a split URL test sends visitors to two different page URLs. Both split traffic at random and compare results.

How long should a landing page split test run?

Until it reaches the sample size you planned before starting, and for at least two full weeks. Full weeks matter because weekday and weekend visitors behave differently. Don’t stop early because one version looks ahead.

What should I split test first on a landing page?

Start with the things that change what people are willing to pay: the offer and price, then the headline and promise, then proof, form length, page length and the guarantee. Leave button colors for when you have a lot of traffic.

How much traffic do I need to split test a landing page?

It depends on your conversion rate and the size of the lift. On a sales page converting at 5%, spotting a lift to 6% takes about 8,200 visitors per version by my math. Smaller lifts need far more: halve the lift and you need four times the traffic.

Should I use a multivariate test instead of an A/B test?

Use an A/B test when you’re changing one thing, or one clear idea. Use a multivariate test when you want to test several elements at once, like headlines, images and buttons, and find the best combination. It needs more traffic, so pair it with a multi-armed bandit split on smaller pages.

Is a multi-armed bandit a type of test?

No. A multi-armed bandit is a way of splitting the traffic, not a different test. You run an A/B or multivariate test and choose an equal, custom or bandit split for it. The bandit sends more visitors to the version earning more while the test runs.

Should I use a split URL test instead of an A/B test?

Use a split URL test when the new version is a whole new page, like a long page against a short one. Use an A/B test on the same URL when you’re changing one part of the page, because it keeps your ads and tracking the same.

My final thoughts

Split testing isn’t hard because of the tools. It’s hard because it’s easy to fool yourself. Stop early, change five things at once, read the wrong metric, and you’ll find a “winner” every single time.

If I were starting on a funnel page today, I’d test the offer or the headline first, decide the sample size before launch, run full weeks, and judge the result on revenue per visitor. On small traffic, I’d only test changes big enough to show up.

That routine is boring, and boring is what makes the results real. When you’re ready to run one with every order, bump and upsell credited to the right version, book a demo.

Test bigger things, and only believe the numbers that reach your bank account.

Built for revenue

Test for revenue, not just clicks.

Run A/B, multivariate, and multi-armed bandit experiments — with real revenue tied to every variation.

We’ll increase your conversion rate by 30% or you don’t pay. Terms apply.

✓

Revenue per variation

✓

A/B & MVT

✓

Multi-armed bandit

Test for revenue, not just clicks.

A/B, multivariate and multi-armed bandit experiments, with real revenue tied to every variation.


Start optimizing for revenue today not just conversion rates.

Run experiments, track revenue per variation, and scale what actually makes you money.

We’ll increase your conversion rate by 30% or you don’t pay. Terms apply.


Start optimizing for revenue today not just conversion rates.

Run experiments, track revenue per variation, and scale what actually makes you money.

We’ll increase your conversion rate by 30% or you don’t pay. Terms apply.


Start optimizing for revenue today not just conversion rates.

Run experiments, track revenue per variation, and scale what actually makes you money.

We’ll increase your conversion rate by 30% or you don’t pay. Terms apply.