# Landing page split testing for presells · Presellbook

> Source: https://presellbook.com/articles/landing-page-split-testing

Testing and measurement

## Split testing a presell page

Change one element per test, write down the sample size and the stopping rule before any traffic arrives, and call the winner on earnings per ad click (EPC) or revenue per presell visitor. Treat presell click-through as a warning sign along the way.

By Presellbook editors. Published 12 October 2026.

This guide draws on published affiliate testing playbooks, experimentation documentation and practitioner threads. It reports no tests of our own (see [how we research guides](/editorial-policy)). Where it gives numbers, they come from the sources linked beside them. Event tracking, click IDs and per-source reports are covered in [how to measure presell-to-sale conversion by traffic source](/articles/presell-conversion-rate), and this page assumes you have them.

### What you are split testing on a presell

A presell is the page between the ad and the offer: an advertorial, a listicle, a quiz or a short bridge page. In a presell split test the order page stays put. You change something on the page before it, usually the headline, the lead, the proof block, the button or the whole layout, and see whether more of the same traffic ends up buying.

Binom's affiliate course sums up the discipline as ["one variable, a pre-declared sample size, and a stop rule written before the test starts"](https://binom.courses/split-testing-landers-and-offers/). One variable, because two changes at once leave you unable to say which one moved the number. If you edit a variant's copy halfway through, the traffic before and after the edit saw different pages, so restart that arm and count from zero.

Keep everything outside the page stable too: the ad creative, the bid, the targeting and the destination URL. Serve both variants behind one tracking URL so the ad never changes. Swapping the URL in a live ad can send it back through review and leaves the two periods on different traffic.

### What to test first on presell pages

Published advice splits into two schools. Binom puts the big decisions ahead of the page: ["The offer, then the angle, then landing page versus direct link"](https://binom.courses/split-testing-landers-and-offers/), with headline tweaks after that. CB-Stats, writing about presells for an offer already chosen, says ["Test the headline first"](https://cb-stats.com/blog/affiliate-ab-testing-landing-pages), along with the primary button, because those move click-through the most.

Both are right at different stages, and the stage tells you which to follow:

* **New offer, or a payout you have not proven.** Test the offer and the angle first, and whether a presell beats sending traffic direct. Polishing a headline on an offer that cannot pay back is wasted spend.
* **Proven offer, weak presell click-through.** Start with the headline, the hook in the first screen and where the call to action sits.
* **Proven offer, healthy click-through, weak sales.** Look at what the page promises against what the checkout delivers: the proof block, the lead's claim, the hand-off copy above the button.

Format is a bigger swing than any headline. If you are deciding between a video and a written presell, that test comes first; [VSL presell or text presell: which to run first](/articles/vsl-vs-text-presell) covers how to frame that test.

### A/B testing an advertorial

On a native advertorial, the first screen has to match the ad's claim before anything else is worth testing. Once it does, the headline and the hero image are the usual first variables. The "Advertisement" label and its placement are a compliance requirement, so treat them as fixed. Only test two placements if both have been cleared as compliant.

Native platforms add a wrinkle: they push delivery toward whichever ad gets early clicks, so two ads pointing at two advertorials rarely get an even share. OpenAdLibrary's fix is ["rotating destinations server-side in your tracker"](https://openadlibrary.com/blog/native-ads-ab-testing/): one ad, one URL, and the tracker sends each visitor to a variant. Our [carbon steel pan advertorial example](/examples/carbon-steel-pan-advertorial) has a teardown that lists what to test first on a page of that shape, and the [advertorials hub](/topics/advertorials) collects the rest.

### How to split the traffic

A 50/50 split gets you a readable answer soonest, so use it when you need a clean read between two pages. A challenger split, such as 80/20 in favour of the current page, makes sense only when the control is already proven and you want to limit what a weak challenger can cost. The price is a longer test, because the small arm takes longer to collect its sample.

Whatever the ratio, set it in the tracker or the page builder's rotation, keep it random, and keep each visitor on the same variant if they come back.

### Pick the metric before you pick the winner

Presell click-through is quick to move and quick to read, which is why it tempts people into early calls. Tracked.it's guidance is ["Do not optimize only for prelander CTR."](https://tracked.it/articles/affiliate-prelander-tracking/) Remoby puts the roles in one line: ["EPC is the scorecard. CTR is the warning light."](https://remoby.com/blog/affiliate-landing-page-strategy-framework/) A headline that lifts clicks while EPC falls is sending curious readers who were never going to buy.

Where payouts differ between offers or the funnel has more than one step, [Binom's course](https://binom.courses/split-testing-landers-and-offers/) makes EPC the comparison, because a higher conversion rate on a lower payout can still earn less. And when one arm has a presell and the other goes direct, the arms have a different number of pages, so OpenAdLibrary scores them on [end-to-end EPC](https://openadlibrary.com/blog/do-pre-landers-increase-conversions/), earnings per ad click.

Split the result by traffic source and sub-ID before comparing variants. A variant can win on native and lose on social, and a blended number hides that.

### How long to run a landing page split test

There is no fixed number of days. The run length falls out of three inputs: your baseline rate, the smallest lift worth detecting and your daily traffic. ROAS365's rule is to ["decide sample size and end date before launch"](https://www.roas365.com/blog/landing-page-ab-test-sample-size-significance-2026), then divide the required sample by daily visitors to get the date.

Set a floor on calendar time as well. Unbounce's duration calculator recommends ["at least one to two weeks"](https://unbounce.com/ab-test-duration-calculator/), and Optimizely's documentation says ["Run tests for at least one business cycle (seven days)."](https://support.optimizely.com/hc/en-us/articles/4410283969165-How-long-to-run-an-experiment) Weekday and weekend buyers behave differently, and a test that only saw Tuesday to Thursday has not met both.

The calculators assume steady traffic. If you raise the bid, swap the creative or add a source mid-test, the traffic after the change is different, and the clean comparison only covers the days before it.

### Presell split test sample size

What you count decides how much you need:

* **Counting clicks to the offer.** CB-Stats suggests ["roughly 200–300 clicks per variation"](https://cb-stats.com/blog/affiliate-ab-testing-landing-pages) before treating a presell click-through result as stable. Treat that as a minimum before you look, and check it against a calculator.
* **Counting sales.** Binom's sample-size table puts a 2% baseline and a 50% relative lift at ["sixty to eighty conversions per variant"](https://binom.courses/split-testing-landers-and-offers/). Smaller lifts need far more.

A thread on AffiliateFix shows why the floor is only a floor. The poster of a CPA lander test ["Pre-committed to 2,500 clicks per side"](https://www.affiliatefix.com/threads/truth-about-most-split-tests-a-b.182958/). The variant that led at around 200 clicks lost once the rest of the traffic came in. It is one forum report, useful as an illustration of how early leads reverse.

With low volume you have two honest choices: run longer, or test only changes big enough to show up in a small sample. Run your own baseline through a sample-size calculator either way.

### When to call a winner, and when to kill the test

Write the stopping rule down before launch, for example a set number of conversions or a set spend per variant, whichever comes first. Binom's rule is to ["run until each variant reaches your pre-declared minimum, not until the result looks good"](https://binom.courses/split-testing-landers-and-offers/). Read the result on the end date you set. Checking every morning and stopping the first time a calculator turns green is how the early leader in the AffiliateFix thread would have been crowned.

Two reasons justify stopping early: clear harm (a broken page, a tracking fault, a compliance problem) or an arm so far behind after the committed sample that more traffic cannot rescue it. Rising spend with no clear answer is not one of them.

Once you have a winner, make it the control, write down the hypothesis and the result, and queue the next single-variable test.

### Mistakes that waste presell tests

* **Changing several things at once.** A new headline, image and button in one variant leaves you guessing which one did the work.
* **Stopping at 100 to 200 clicks because one arm leads.** Early leads reverse.
* **Optimising click-through while EPC falls.** The presell looks better each week and earns less.
* **Editing a variant mid-test without restarting it.** The arm now holds two pages' worth of data.
* **Blending traffic sources in one comparison.** Segment first. The [presell conversion guide](/articles/presell-conversion-rate) sets out the per-source report.

### What to run next

Once the page itself is settled, the next tests are about format and steps. For video against text, read [VSL presell or text presell](/articles/vsl-vs-text-presell). For a multi-step presell, [how to build a quiz funnel for paid traffic](/articles/quiz-funnel) covers the result-page events. For the shortest presell there is, see [what to put on a bridge page for affiliate offers](/articles/bridge-page-affiliate-marketing). The [testing and measurement hub](/topics/testing-and-measurement) gathers the rest, or browse [all guides](/articles).

---

[ ](/) 

Working presell pages, advertorials and order pages you can copy as clean HTML, each with a teardown of why it is built that way, plus guides on direct-response copy, compliance and testing.

[Write to the editors](/contact)

### Read

* [Example pages](/examples)
* [Guides](/articles)
* [All topics](/topics)
* [Search](/search)

### About us

* [About](/about)
* [Authors](/authors)
* [Editorial policy](/editorial-policy)
* [How we make money](/how-we-make-money)

### Legal

* [Terms](/terms)
* [Privacy](/privacy)
* [Contact](/contact)

© 2026 Presellbook. Some links to tools are affiliate links: if you buy through one we may earn a commission, at no cost to you. [How we make money](/how-we-make-money).