Landing Page A/B Testing: A Practical Guide for UK Teams

Fabio Embaló

Co-founder & CEO, Viaduct Generation

Published

November 9, 2026

Landing page A/B testing, also called split testing, means showing two or more versions of the same page to different visitors at the same time and measuring which one delivers more of a commercial outcome you defined in advance. Nielsen Norman Group describes it as a quantitative method that compares live design variations against predetermined business-success metrics, with traffic divided so that each visitor sees only one variation. Done properly, it replaces opinion with evidence. Done carelessly, it produces confident answers to questions the data never actually settled.

This guide sets out how we approach it: designing the test, sizing it honestly, choosing between A/B and multivariate methods, prioritising a roadmap and reading the result.

Viaductgen
Turn Testing Into Measurable Growth
Viaductgen connects conversion rate optimisation with search, brand, and performance in an AI-powered growth system focused on commercial outcomes.
Explore our growth services

Table of Contents

How landing page split testing works

A landing page has one job: to move a visitor towards a single action, whether a demo request, a purchase or a donation. A split test isolates one change to that page, sends a share of traffic to each version and compares the outcome. Because both versions run concurrently, seasonality, campaign changes and news cycles affect each version equally, which is the structural advantage over comparing this month with last month.

In practice, platforms such as HubSpot’s page testing serve both variants from the same live URL and keep a visitor on a consistent version within a session, so the experience stays coherent. Most tools default to an even split, though uneven splits are possible when a team wants to limit exposure to a risky variant.

The method answers “which version performed better”. It does not tell you why, which is why the quality of your hypothesis matters more than the choice of software.

Write the hypothesis before you build the variant

Every test we run begins with a written hypothesis that links evidence to a change and a measurable consequence. A workable structure is:

  1. Evidence: what you have observed (analytics drop-off, session recordings, sales call objections, search term intent).
  2. Change: the single alteration you will make.
  3. Audience: who sees it, for example paid search traffic on mobile.
  4. Primary metric and threshold: what should move, and by how much to be worth acting on.
  5. Falsifier: the result that would show you were wrong.

A vague hypothesis such as “a cleaner design will convert better” cannot lose, so it cannot teach you anything. A precise one can.

Choose one primary metric, as close to revenue as your volume allows. For a B2B lead generation page, qualified leads are a better measure than raw form submissions, because a shorter form can lift submissions while lowering lead quality. For an e-commerce page, revenue per visitor usually tells you more than click-through. For a charity, completed donations and average gift matter more than button clicks. Add two or three guardrail metrics, such as bounce rate or refund rate, that must not deteriorate while the primary metric improves.

What to test on a landing page

Start where the potential effect is largest, which is almost always the message before the layout.

  • Headlines: benefit-led versus feature-led, short versus long, and message match, meaning whether the headline echoes the advert or search query that brought the visitor.
  • Calls to action: wording that emphasises the value of the action rather than the action itself, placement above or below the fold, and a single prominent button versus repeated placements.
  • Forms: the number of fields, the order of questions and whether a phone number is mandatory. Pair this with a lead-quality measure.
  • Proof: the position and type of testimonials, client logos, case study snippets and credentials.
  • Offer framing: a demo versus an audit, a free trial versus a guided walkthrough. This is often the highest-impact variable and the one teams neglect.
  • Media: hero image versus product screenshot versus short video.

HubSpot’s documentation lists calls to action, copy, media and form length as practical variables, which matches what we see across B2B and e-commerce accounts. If your messaging itself is unclear, fix that first: our piece on B2B website messaging built on buyer phrases shows how to find the language worth testing.

How much traffic you actually need

This is where most landing page programmes fail quietly. A test needs enough visitors in each variant to detect the size of effect you care about, and small effects need far more traffic than large ones. Practitioners commonly plan for 95% confidence and 80% power, estimate the sample size before launch, and run for at least one full business cycle so that weekday and weekend behaviour are both represented.

The practical consequences are these:

  • If a page receives a few hundred visits a month, test bold changes, such as a different offer or a rebuilt page structure, rather than button colours.
  • If the required duration stretches to months, do not run the test. Test a bigger change, move the metric up the funnel (for example, from purchase to add-to-basket), or use qualitative evidence such as user interviews and sales call feedback.
  • Fix the stopping rule in advance. Looking at results daily and stopping the moment one variant leads inflates false winners and exaggerates effect sizes.

We would rather run one properly powered test than four underpowered ones. An inconclusive result you trust is more useful than a win you cannot reproduce.

A/B testing versus multivariate testing

An A/B test compares whole versions, which can differ in many ways. A multivariate test changes several elements at once in a structured set of combinations, so it can attribute effect to individual elements and their interactions.

A/B and multivariate testing compared

The trade-off is traffic. Nielsen Norman Group explains that multivariate testing needs much more traffic and suits incremental refinement, whereas A/B testing suits dramatic redesigns. Its recommended sequence for major changes is to A/B test the original against the new design, then refine the winner with multivariate tests.

The arithmetic is unforgiving. Two headlines, two images and two calls to action produce eight combinations, each sharing the same traffic. Shopify UK suggests that A/B tests usually return reliable results in one to two weeks, and offers a rule of thumb that traffic doubles with each variant added: 10,000 sessions for two variations, 20,000 for three and 40,000 for four. Matomo recommends aiming for 10,000 or more visitors a month before considering multivariate.

Situation Better method
Under about 10,000 monthly visitors to the page Sequential A/B tests
Testing a new layout or a different offer A/B test
Refining several elements on a proven page Multivariate, with ample traffic
Unsure which element drives performance A/B test one element at a time, in priority order

For most mid-market landing pages, a disciplined sequence of A/B tests is the more reliable route.

Build a testing roadmap, not a list of ideas

A backlog of ideas becomes a strategy once each item is scored, scheduled and documented. Two scoring models are widely used: ICE (impact, confidence and ease) and PIE (potential, importance and ease), each element scored from 1 to 10 and combined into a ranking. The model matters less than applying it consistently and tying confidence to evidence rather than enthusiasm.

A roadmap should go beyond ranking. Optimizely’s guidance on experimentation roadmaps recommends tracking the experiments in priority order, a balanced schedule, timelines, potential interference with other campaigns, who is involved, available resources and alignment with company objectives. That last point is often missed: a test that wins on conversion but conflicts with a brand repositioning is not a win.

In practice, we recommend:

  1. A ranked backlog with the hypothesis, metric and expected effect recorded for each item.
  2. A fixed cadence, even if that is one well-documented test a month for lower-traffic pages.
  3. No overlapping tests that share the same audience and page.
  4. A quarterly review to re-score the backlog using what you have learned.

We have set out how to link the calendar to commercial outcomes in building a CRO testing calendar that lifts revenue. If your tests depend on advertising creative, the 90-day creative testing framework shows how page and ad tests can reinforce each other. For teams wondering where machine learning fits, see our view on AI-driven A/B testing.

If you are still at the stage of setting up the page itself, a guide to building a landing page with the right tracking and forms in place makes later experiments far easier to measure.

Reading the result

A statistically significant result is not automatically a commercially useful one. Before launch, set the smallest improvement that would justify the cost and risk of rolling out the change, and judge the result against that threshold and your guardrail metrics, not against significance alone. Changing the metric, the conversion window or the stopping rule after seeing the data invalidates the test.

When a test finishes, record four things: the hypothesis, the result with its uncertainty, the decision taken and what the result suggests for the next test. A win on a headline might point towards testing the same message in your adverts. A loss on a shorter form might show that your audience values the reassurance of a fuller qualification step. The compounding value of testing lies in this record, not in any single uplift.

Where Viaduct Generation fits

Our conversion rate optimisation service covers funnel mapping, A/B and multivariate testing and the analysis that connects page changes to revenue. It runs inside our 90-day growth sprint, so testing is planned alongside search, paid media and email rather than treated as a separate exercise. We work with B2B businesses, e-commerce companies and charities that want a structured programme with defined deliverables and commercial targets.

FAQ

How long should a landing page A/B test run?

Long enough to reach the sample size you calculated before launch, and for at least one full business cycle, which is often a minimum of two weeks. Stopping early because one version looks ahead raises the risk of a false winner. If the required duration is unrealistic, test a bolder change instead.

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

In most marketing contexts the terms mean the same thing: comparing versions of a page by splitting traffic between them. Some teams use “split test” for tests that send traffic to entirely separate URLs, but the method and the statistics are the same.

Can I run a landing page test with low traffic?

Yes, but adjust the approach. Test large changes such as a different offer, use a metric higher up the funnel, and run one documented test at a time. Where the traffic still cannot support a result, qualitative evidence from user interviews and sales calls is more reliable than an underpowered test.

Should I test one element or several at once?

Test one change at a time unless you have the traffic for multivariate testing. A single-variable test tells you which change caused the effect, and it reaches a conclusion faster on typical landing page volumes.

Sources

About the Author

Fabio Embaló

Co-founder & CEO, Viaduct Generation

Fabio co-founded Viaduct Generation in 2020 with a belief that the gap between agency output and business impact was structural, not incidental. He leads the agency's strategic direction, client partnerships, and the development of the Growth Engine methodology. With a background spanning organic search, content strategy, and digital transformation, he has spent his career building systems that connect digital activity to commercial outcomes.

AI Strategy Growth Architecture SEO & AEO Client Partnerships