Landing pages for Meta ads

How do I A/B test a landing page for Facebook ads?

How to A/B test a landing page for Facebook ads: split traffic from the same ads, change one element per variant, and judge on cost per lead.

Published

A/B test a landing page for Facebook ads by sending traffic from the same ads to two variants at the same time, changing one element per variant, and judging the winner on the cost per lead from the leads you actually received. Keep the creative fixed so the landing page is the only difference, and write your cut and winner rules before you spend anything.

What you are really comparing: the creative and landing page pair

A creative makes a promise in the feed, and the landing page either keeps that promise or breaks it. That is why the same landing page can produce cheap leads behind one ad and expensive ones behind another: each ad brings a visitor with a specific expectation.

Test landing page variants behind your single best ad and you learn how those variants behave for one kind of visitor, and nothing about the rest. Report cost per lead per pair of creative and landing page, and keep the pair visible when you decide what to scale. A Meta ads creative testing framework for lead gen sets out the order that keeps these reads comparable from one round to the next.

Fixed 50/50 split or adaptive allocation

A fixed split gives each variant half the traffic until it reaches a pre-computed sample size. It is easy to defend, and it carries a cost that is easy to forget: for the whole run, half your paid clicks go to the variant that is losing. You also cannot stop early when the numbers look good, because peeking and stopping early inflates the false positive rate.

Adaptive allocation, run with a bandit rule such as Thompson Sampling, works differently. Each variant keeps a belief about its conversion rate, the tool serves the variant that currently looks better more often, and it keeps sending a share of traffic to the others so an early lucky streak does not freeze the test. Allocation concentrates as evidence builds, so the money lost on a clearly worse variant shrinks within days instead of lasting until the end.

Fixed 50/50 splitAdaptive allocation
Traffic spliteven until the sample size is reachedadjusts with every visitor
Cost of a clear loserpaid for the whole testshrinks quickly
Stoppingpre-computed, no early stopno stopping rule needed
Reading the test mid-runinvalidmeaningful, as a win probability
Best fita decision you must defend statisticallyongoing optimisation

How to run the test, step by step

  1. Choose the variable. Change one element per variant: the offer, the form, the first screen, the proof or the call to action. Two changes in one variant make the answer unreadable.
  2. Keep everything else fixed. Same ads, same audience, same optimisation event, same budget per variant. If the creative changes too, you are testing a pair, which is a different question with a bigger sample requirement.
  3. Run the variants at the same time. Traffic on Monday and traffic on Friday are different populations, so a week by week comparison measures the calendar as much as the landing page.
  4. Count the result yourself. A lead is a form your server received, not a number a platform reports. That count is the one you can divide spend by without an argument about attribution, and landing page views vs link clicks explains why platform numbers drift apart from visits.
  5. Size the spend in multiples of your target cost per lead. Around three times the target screens a variant that produced nothing, ten times ranks two variants against each other, and twenty-five to thirty times, on finalists only, is what a winner needs at lead generation volumes.
  6. Write the decision rules before launch: what gets cut, what gets more budget, what counts as a win. Changing the rules after you see the data is how noise turns into strategy.
  7. Confirm on qualified leads. A variant with more leads at a lower cost per lead that fills the CRM with people who never answer has not won.

Mistakes that make a landing page test answer nothing

  • Changing more than one thing per variant, so no result can be attributed to an element.
  • Comparing two periods instead of two variants.
  • Calling a winner on a handful of leads. At small counts, the gap between two identical variants is wide often enough to crown the worse one.
  • Testing something that cannot move the number, such as the colour of a button below a form most visitors never reach.
  • Optimising conversion rate instead of cost per qualified lead. A landing page converting at 30% with unqualified leads is worse than one converting at 8% with the right ones. What a lead generation landing page needs to include covers the elements that decide which of the two you get.

What to test first on a lead generation landing page

Roughly in order of how much the result can move:

  1. The offer: what the visitor gets, and how it is framed.
  2. The form: how many fields, in what order, and what is asked first.
  3. The first screen: whether the offer is clear within a few seconds on a phone.
  4. The proof: numbers, names and reviews a visitor could check for themselves.
  5. The call to action: the verb, the placement, and how many competing actions sit beside it.

What a landing page A/B testing tool has to do

  • Split traffic when the landing page is served, before any redirect, so the click identifier survives the split.
  • Log the click on the domain that serves the landing page, so every lead carries the ad and the variant that produced it.
  • Read results per variant and per creative and landing page pair, because an average across ads hides the interaction.
  • Adapt allocation as leads arrive, and show a win probability rather than a p-value you have to interpret.
  • Keep versions of each variant, so an edit that broke the tracking can be reverted.

DropCPL allocates traffic between landing page variants with Thompson Sampling and measures the real cost per lead of every creative and landing page combination. The click is logged first-party on the landing page domain through an HttpOnly cookie that holds the click_id, with no redirect in the path, so a lead stays tied to its ad, its creative and its landing page even when the browser blocks the Pixel. The allocation rule itself is explained in A/B testing landing pages.

Questions on this topic

All questions on this topic →

FAQ

Can I run a landing page test with Meta's own A/B test tool?+

Meta's A/B test compares versions of a campaign, an ad set or an ad, shows each version to a separate segment of the audience so nobody sees both, and ranks them on cost per result. That works well when the two versions point at different landing pages. When you want two landing page variants behind the same ad, count the leads on your own domain instead.

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

Long enough for each variant to collect the spend multiples you set before launch, which at lead generation volumes usually means weeks rather than days. A fixed split has to reach its sample size before you read it. An adaptive test can be read at any point, as a win probability, without invalidating the result.

How many landing page variants should I test at once?+

Two is the practical starting point. Every extra variant spreads the same traffic thinner and pushes a readable answer further away. Add a third only when one variant is clearly ahead and the volume per variant still meets your threshold.

Does the ad creative have to stay the same during the test?+

Yes if the question is which landing page is better. Changing the creative and the landing page together answers a different question, and a more useful one for scaling, but each combination then needs its own spend before the cell means anything.

Ready to find out what
converts, for real?

Free to start.

Start free →