A/B testing a landing page means showing two versions to comparable visitors at the same time and keeping the one that produces more of the outcome you care about. For lead generation pages, that outcome is qualified leads, not clicks. The method is simple; the discipline is in testing one change at a time, deciding in advance how long the test runs, and refusing to call a winner early.
Most B2B landing pages get too little traffic to test small tweaks like button colors. They can still test the changes that matter most: what the page offers, when it asks for an email, and what the headline promises.
This guide covers how to set up a test, how to think about sample size without guessing, what to test first, and when you should skip the test and simply fix the page.
Set up the test before you build the variant
Write the test down in five lines before anyone designs anything:
- The change. One specific difference between version A and version B.
- The reason. What you saw that suggests it will help: a drop in the data, a pattern in recordings, something prospects said on calls.
- The primary metric. The single number that decides the winner, such as leads per visitor or qualified leads per visitor.
- The duration. A fixed number of full weeks, set from the sample size thinking below.
- The decision rule. What result means you keep B, what means you keep A, and what you do if it is unclear.
Split traffic randomly, at the same time, from the same sources. Testing version A in March and version B in April compares two months, not two pages.
Change one thing at a time
If version B has a new headline, a shorter form and different proof, and it wins, you do not know which change did it. Worse, one change may have helped while another hurt, and the combined result hides both.
One change does not have to mean one small change. "Show the result before asking for the email" is one change, even though it moves several elements. The test is clean as long as every difference between A and B serves the same idea.
When you want to compare a full redesign with the current page, treat it as a decision, not a learning test. You will find out which page performs better, but not why. Then go back to single-change tests on the winner.
Think about sample size before you start
How many visitors a test needs depends on two things: your current conversion rate and the size of difference you want to be able to detect. A large difference shows up with fewer visitors. A small difference needs many more.
Work it through before you launch:
- Find your baseline. Take the page's leads divided by visitors over a recent stable period.
- Decide the smallest difference worth acting on. If B would need a large lift to justify the effort, you only need to detect a large lift.
- Use a sample size calculator. Enter the baseline and the difference. It tells you the visitors needed per version.
- Divide by your weekly traffic to get the duration, then round up to full weeks so every weekday is represented equally.
If the answer is longer than about two months, the test is impractical. Either test a bolder change that would produce a larger difference, or skip testing and make the change on judgment, then watch the numbers.
Two rules protect you from false winners:
- Do not stop when the dashboard first shows significance. Results swing early. Run to the planned duration.
- Do not extend a test because you dislike the result. Decide the duration once.
What to test first
Start with changes closest to the conversion and largest in effect. In rough priority order for lead generation pages:
1. When you ask for contact details
This is the change with the most room on most pages. In a calculator funnel we audited for a client, 59 to 63% of visitors left at a form placed before the result. On our own sites, ten Foxyapps tools that asked for free text and an email before showing anything were loaded about 8,500 times over about two years and produced 21 leads, about 0.25% (ViralSales inventory, November 2026). Test showing the answer first and asking for the email after.
2. The offer itself
A demo request against a short check with an instant result. A PDF guide against a scorecard. Different offers attract different visitors, so track qualified leads, not only total leads.
3. The headline's promise
Test promises, not wording. "See where your funnel leaks in three minutes" against "Book a funnel review" tests two different offers to the reader. Synonym swaps rarely move anything you can detect at B2B traffic levels.
4. Form fields and question format
Fewer fields, or multiple choice in place of free text. The multi-step form guide covers when splitting a form across screens helps.
5. Proof placement
A testimonial beside the main claim against the same testimonial lower on the page. See social proof examples for which proof to use.
Button colors, icon styles and small layout shifts come last, if at all.
When not to run a test
A test is the wrong tool in several common situations:
- Something is broken. A form that drops a field, a button hidden on phones, a missing confirmation message. Fix it. The CRO checklist finds most of these in an afternoon.
- Traffic is too low for any meaningful difference to show up within two months. Make the change on judgment and monitor.
- You cannot measure the outcome reliably. If analytics and the CRM disagree, fix tracking first.
- The page is in an unusual period: a launch, a big campaign, a holiday, press coverage. The visitors are not typical.
- The change is required anyway, such as a legal notice or an accessibility fix.
Make sure the measurement is sound
A test is only as good as its counts. Before launching, check three things:
- Each step is counted. Views, starts, completions and leads for each version, so you can see where B differs from A. A ViralSales check counts these separately for this reason.
- Tracking is not blocked. Ad blockers match on words in request paths. Counted on 16 November 2026 across EasyList and related lists, 1,410 rules match
/track, 318 match/eventsand 292 match/collect. If blocked visitors skew toward one version, the result is wrong. - Leads reach the CRM with the version recorded, so you can judge lead quality weeks later, not only volume.
On WordPress sites, WP Commit runs scheduled tracking checks that catch a tag that stopped firing midway through a test.
Record every result
Keep a simple log: date, page, change, reason, duration, visitors per version, result, decision. Include tests that showed no difference. Over a year, the log tells you what your buyers respond to, and it stops the team from rerunning the same idea. Ideas for the queue can come from a CRO audit or from an outside review such as Behavly, which suggests copy and layout changes.
Where to start
- Pick your highest-traffic lead page, write down its baseline leads per visitor, and work out with a calculator how many weeks a test would need.
- If the duration is under two months, write the five-line plan for one test of when you ask for the email.
- Fix anything broken on the page before the test starts, and confirm form submissions match between analytics and the CRM.