Landing Page Testing: What to Test First (and What to Skip)

Updated 4 November 2026

What should I A/B test first on a landing page?

Test what most visitors see and act on first: the headline, the call to action and the offer. Then test the form, proof, hero visual and page length. Skip button colours and small wording changes until those are settled. Each test needs one hypothesis tied to evidence, and enough conversions to give a reliable answer.

Most landing page testing goes wrong before the test starts. Teams test what is easy to change, not what is likely to matter, or run tests on pages that will never get enough conversions to give a clear result. This guide gives you a starting backlog in priority order, a quick way to check whether your traffic can support a test at all, and a short list of tools with prices checked on 4 November 2026. If you have not looked at the page closely yet, start with a landing page audit: it gives you the evidence your hypotheses need.

Key Takeaways

• Prioritise tests by how many visitors see the element, how strong your evidence is, and how cheap the change is. The headline, call to action and offer usually come first.
• Write each test as a hypothesis: "Because we saw X, we expect changing Y to improve Z." No evidence, no test.
• At a 3% conversion rate, detecting a 20% relative lift needs about 13,000 visitors per version, using Evan Miller's sample size formula at a 5% significance level and 80% power.
• If a page cannot reach that sample in a few weeks, fix clear problems directly instead of testing them.
• GrowthBook and PostHog both run tests for free, but GrowthBook's visual editor is a paid Pro feature, and PostHog warns its no-code changes can be overwritten on pages that re-render. Landing page builders like Unbounce include A/B testing on their mid-tier plans.

Jump to: the first 7 tests · how to prioritise · what to skip · traffic you need · testing tools · a testing strategy.

The first 7 landing page tests, in order

This order is editorial judgement, not a measured ranking. It puts the elements every visitor sees and acts on ahead of the elements only some visitors reach. Change the order if your evidence points elsewhere: if recordings show people abandoning the form, the form goes first.

1. Headline that matches the ad or search query. Hypothesis: visitors who see the promise from the ad or search result repeated in the headline will stay and read on, because they know they are in the right place. Evidence to look for: high bounce from paid traffic, or a headline that describes the company rather than the offer.
2. Call-to-action wording and position. Hypothesis: a button that names what the visitor gets ("Get my quote", "Start free trial"), placed where the first screen ends on mobile, will be clicked more than a generic "Submit" placed lower down. Evidence: low click-through to the form or checkout, or a call to action missing from the first screen.
3. The offer itself. Hypothesis: a more specific or lower-risk offer (a free trial instead of a demo, a price instead of "contact us", a guarantee) will turn more interested visitors into leads or buyers. Evidence: visitors reach the call to action but do not click, or sales calls show the same objection again and again.
4. Form length. Hypothesis: removing fields you can collect later will increase completions. Nielsen Norman Group's form design guidance recommends removing fields that can be derived, collected later or omitted, and reports cutting one of its own forms from six fields to two. Evidence: form starts are much higher than form submissions.
5. Proof next to the decision. Hypothesis: moving reviews, client logos or a specific result next to the call to action will reassure visitors at the point of decision. Evidence: proof exists on the page but sits far from the button, or visitors leave from the pricing or form section.
6. Hero visual. Hypothesis: showing the product, the result or a real customer will explain the offer faster than a stock photo or abstract graphic. Evidence: visitors scroll past the hero without engaging, or survey answers show people are unsure what you sell.
7. Page length and section order. Hypothesis: for a high-commitment offer, a longer page that answers objections will convert better; for a simple offer, a shorter page will. Evidence: scroll maps show visitors stopping before key sections, or the page repeats itself.

How do you prioritise landing page tests?

Score each idea on three things, then sort: how many visitors see the element (reach), how strong the evidence behind it is, and how cheap it is to build. An idea that every visitor sees, with recordings or survey answers behind it, and a one-hour build beats a clever redesign of the footer.

Nielsen Norman Group's A/B testing guide explains why the evidence step matters. It notes, citing VWO data, that only one in every seven A/B tests is a winning test, and that the rate is likely lower without a strong, data-based hypothesis. Each test also uses up weeks of traffic, so every weak idea you run delays a strong one.

A useful hypothesis has three parts:

1. What you saw. "Recordings show most mobile visitors never scroll to the form."
2. What you will change. "Add a call to action to the first screen that jumps to the form."
3. What you expect to move. "Form submissions per visitor will rise, with no fall in lead quality."

The last part needs a guardrail metric as well as a main metric. NN/g recommends tracking guardrail metrics alongside the main one, so that a change that lifts clicks but hurts sales or retention is not called a win.

What should you skip testing?

Skip anything that is broken, anything too small to detect, and anything you cannot learn from. Testing these wastes the limited traffic you have.

On multivariate testing, NN/g's comparison with A/B testing notes that two variables with two variants each already produce four versions to compare, and that more traffic is typically required to reach statistical significance. Most landing pages are better served by one clear A/B test at a time.

How much traffic do you need to A/B test a landing page?

You need enough visitors for each version to detect the smallest lift worth acting on. At a 3% conversion rate, detecting a 20% relative lift (3.0% to 3.6%) takes about 13,000 visitors per version, or 26,000 in total for a simple A/B test. A page with 1,000 visitors a week would need about six months.

The figures below come from the formula behind Evan Miller's sample size calculator, using its default settings of a 5% significance level and 80% power, for a two-sided test. They are visitors per version; double them to get the total across the original and one variant.

Source: calculated on 4 November 2026 with the sample size formula published in Evan Miller's calculator, at α = 0.05 and power = 0.80. NN/g's A/B testing guide reaches the same order of magnitude for its own example: a 3% baseline and a 20% minimum detectable effect need about 13,000 users.

Two rules follow from this table:

• Decide the sample size before you start, and do not stop early. Evan Miller's How Not to Run an A/B Test shows that checking results repeatedly and stopping at the first significant result can push the false-positive rate from 5% to 26.1% in a worst case. NN/g also recommends running a test for at least one to two weeks to cover normal swings in behaviour.
• If you cannot reach the sample in a reasonable time, do not test small changes. Either test a bolder change, which needs fewer visitors, or make the change based on evidence and watch the before-and-after numbers. Our A/B testing best practices guide covers how to plan tests once you have the traffic.

What are the best landing page testing tools?

The best tool depends on who builds your pages and how much you want to pay. If marketers build pages in a landing page builder, use its built-in testing. If developers build the site, a free experimentation platform such as GrowthBook or PostHog can do the job. Teams that want a managed platform with vendor support can get quotes from VWO (now Wingify) or Optimizely.

Sources: each vendor's pricing or documentation page linked in the table, plus GrowthBook pricing and PostHog's no-code web experiments documentation, retrieved 4 November 2026. Unbounce prices are in US dollars. For a wider list, including behaviour analytics tools, see our guide to the best conversion rate optimization tools, and our Hotjar vs VWO vs Optimizely comparison for those three in more depth.

Whatever you choose, run an A/A test first. NN/g suggests testing two identical versions against each other before your first real test: the result should be inconclusive, and if it is not, something in the setup is wrong.

What does a good landing page testing strategy look like?

A good strategy is a loop: find problems, fix the obvious ones, test the uncertain ones, then record what you learned. Testing is the slowest step, so it should only get the ideas that are both important and uncertain.

1. Audit the page. List the problems a visitor would hit, from message match to form friction. The landing page optimization best practices guide is a good checklist.
2. Fix what is clearly broken. Errors, missing calls to action and slow images do not need a test.
3. Turn the rest into hypotheses using the three-part format above, and score them for reach, evidence and cost.
4. Check the traffic for the top idea against the sample size table. If it will not finish in a few weeks, make the change bolder or move to the next page.
5. Run one test at a time per page, with a fixed sample size, a main metric and a guardrail metric.
6. Write down the result, including losses and inconclusive tests. A test that shows no difference still tells you that element is not your bottleneck.

Find your first test

The quickest way to fill the top of your backlog is to see the page through a visitor's eyes. Run a Croopt audit on your page to get a prioritised list of conversion issues, then fix the clear ones and test the rest.

Sources and author

• Nielsen Norman Group, A/B Testing 101 (Tim Neusesser, 30 August 2024), retrieved 4 November 2026. Used for the one-in-seven figure (which NN/g attributes to VWO), the 13,000-user example, the one-to-two-week minimum, guardrail metrics and A/A tests.
• Nielsen Norman Group, Multivariate vs. A/B Testing, retrieved 4 November 2026. Used for the traffic needs of multivariate tests.
• Nielsen Norman Group, Website Forms Usability: Top 10 Recommendations, retrieved 4 November 2026. Used for the field-removal advice and the six-to-two example.
• Evan Miller, Sample Size Calculator, retrieved 4 November 2026. Its published formula and default settings were used to calculate the sample size table.
• Evan Miller, How Not to Run an A/B Test, retrieved 4 November 2026. Used for the 26.1% false-positive example.
• GrowthBook, homepage and pricing; PostHog, pricing and no-code web experiments; Unbounce, pricing; Wingify, pricing; Optimizely, plans. All retrieved 4 November 2026 and used for the tool table.

Kieron Bertani is the founder of Croopt, the landing page audit tool mentioned in this article, and writes about conversion rate optimisation. This article and its hero illustration were created with AI assistance. The test order and the "what to skip" list are editorial judgement based on the sources above, not results from Croopt's own experiments. For corrections or product questions, contact Croopt support.