How-to guides

A/B Testing: A Step-by-Step Guide for Better Conversions

A practical, step-by-step guide to A/B testing. Learn how to plan a test, build variants, read results, and avoid common mistakes.

Published Updated

A/B Testing: A Step-by-Step Guide for Better Conversions. A staircase of dots with 7 steps, one for each step in the guide.

A/B testing shows whether a change to your page actually helps. You show one version to half your visitors and another to the other half, then measure which one gets more clicks, signups, or sales. This guide takes you from picking a test idea to knowing when you can trust the result.

Many teams wait too long to start or change too much at once and can’t tell what worked. Here’s the full process.

The short answer

A/B testing compares two versions of a single page element with real visitors. You split your traffic, usually 50/50, and track one goal like form submissions or button clicks. When one version clearly wins, you keep it and start your next test. Done this way, A/B testing cuts guesswork from your decisions.

You don’t need a huge audience. You do need a clear question, enough traffic to get a reliable result, and the patience to wait for the data.

Step by step

1. Choose one thing to test

Pick a single element on your page. Common candidates:

  • The headline
  • The call-to-action text
  • The button color or placement
  • The lead form length
  • The hero image
  • The offer itself

If you change a headline, a button, and a form in one test, you won’t know which change caused the result. Test one variable at a time.

Start with the element closest to the visitor’s decision. Often that’s the headline or the call-to-action. Both sit right in the path to conversion.

2. Write a clear hypothesis

A test without a hypothesis is a coin flip with extra steps. Write what you expect and why.

A weak hypothesis: “I want to see if this button works better.”

A strong hypothesis: “Changing the button text from ‘Submit’ to ‘Get my free guide’ will increase form submissions because it restates the offer at the moment of action.”

Writing the reason forces you to think about your visitor’s motivation. That leads to better tests.

3. Create the two versions

Version A is your control, the page you have now. Version B is the variant with your one change.

Build both in your landing page builder. If you use Leadpages, duplicate your existing page and make the single edit for the B version. See how A/B testing works in Leadpages.

Keep everything else identical. Same traffic sources, offer, form, and timing. Otherwise a difference might come from outside the test.

Need help with headlines or calls to action before you build the variant? Try the free headline analyzer or use the call-to-action generator for a starting point.

4. Split your traffic

Set up the test so both versions receive traffic at the same time from the same sources. A 50/50 split is standard because it gives each version the same chance.

Don’t show version A this week and version B next week. Seasonal changes, promotions, or shifting ad budgets will skew the result. Compare them simultaneously.

5. Pick one primary metric

Choose a single number before the test starts. For most pages, that is:

  • Form submissions
  • Button click-through rate
  • Purchase completions
  • Email signups

Watching ten metrics invites cherry-picking. You’ll convince yourself a test won because one side number ticked up. Pick the goal that matches the page’s job, and judge the test on that number alone.

6. Let the test run long enough

The biggest mistake is stopping early. You see a clear leader after two days and declare victory. Early data is noisy. Small samples swing wildly.

Run the test until you reach statistical significance. That’s the point where the result is unlikely to be random. Most testers use a 95% confidence level.

You can calculate this in a spreadsheet, or use the free A/B test significance calculator. Enter visitor counts and conversions for each version, and it tells you whether the result is reliable.

There’s no universal minimum time. It depends on your traffic and conversion rate. A page with 10,000 visitors a week reaches significance faster than one with 500.

7. Read the result honestly

If B beats A with 95% confidence or higher, keep it. If not, keep the original.

A test that shows “no significant difference” isn’t a failure. You learned the change doesn’t matter, so you can stop wondering and test something else.

If B loses, revisit your hypothesis. What did you assume about your visitors that wasn’t true? Use that to shape your next test.

Common problems

Stopping the test too early

You check the dashboard on day two. Version B leads by a noticeable margin. You stop the test and make B the permanent page. Over the next two weeks, the numbers even out and B’s win disappears.

Early visitors often aren’t representative. Maybe the first wave came from your most engaged subscribers or from a different ad. Waiting for statistical significance protects you from this.

Testing too many things at once

You rewrite the headline, change the button color, shorten the form, and swap the image. The B version wins, and you don’t know which change did it.

This makes every later decision harder. You can’t carry the lesson forward because the lesson is a jumble. Test one element at a time. It feels slower. It creates usable knowledge.

Tiny sample sizes

A page with 200 visitors a month will take a long time to produce a reliable test. Split into two groups of 100, one extra conversion swings the result by a lot.

If traffic is low, test bigger changes that could produce larger effects, like a new offer or layout. Small button-text tweaks won’t move enough data to read. You can also increase traffic first, through paid ads or better search distribution, before focusing on optimization.

Ignoring outside variables

Your test runs during a flash sale. Or a holiday weekend. Or the week your ad account has a billing issue. Any of these changes who arrives and what they do.

If a major outside event happens during your test, pause it and restart later. The data is contaminated. Better to run a clean test than analyze a messy one.

Testing elements that don’t matter

You spend three weeks testing button color from blue to green. Button color rarely changes conversion on its own because color usually isn’t why someone hesitates.

Test what addresses real friction or motivation: the headline, the offer, the form fields, the page layout. If the visitor isn’t clear on what they get, no button color fixes that. If they don’t trust the price, no headline fixes that alone. Match the test to the real decision in the visitor’s mind.

Not running tests continuously

A single A/B test is a project. A program of A/B tests is a system. Teams that get the most from testing never stop. Each result raises a new question, and each question becomes the next test.

If you treat testing as a one-time cleanup, you’ll make a few improvements and plateau. If you treat it as an ongoing habit, the gains compound.

What to do next

Run your first A/B test this week. Pick one page with steady traffic. Choose the single element most likely to affect conversion, write a hypothesis, build the B version, and launch the test at a 50/50 split.

While you wait for results, two useful steps:

  1. Score your current page. The free landing page analyzer reviews your page and points out practical improvements. Run it on your control before you test. It may surface a bigger opportunity than you planned.
  2. Look at proven page structures. The landing page template gallery shows layouts that work across industries. If your page has a structural problem, no single-element test will fix it. Start from a template built for conversion, then test from a strong base.

When your test reaches 95% confidence, record the result. Keep a simple log: the hypothesis, traffic split, outcome, and the next question it raises. That log becomes your testing roadmap.

A/B testing isn’t complicated. It’s a method. Decide what to test, change one thing, split traffic, wait for real data, and keep what wins. To start building and testing pages with a tool that handles the split for you, start a free trial of Leadpages.

FAQ

Questions, answered

Run it until you reach statistical significance at a 95% confidence level or higher. The exact time depends on your traffic volume and conversion rate. Avoid stopping early just because one version leads after a day or two.

A 50/50 split is the standard. It gives both versions the same exposure, which makes the comparison fair. Uneven splits are rare and usually used only in specific cases like testing on a small segment before a full rollout.

There is no fixed number. It depends on your current conversion rate and the size of the improvement you expect. Higher traffic reaches significance faster. Pages with very low traffic are better off testing larger, higher-impact changes.

Start with the headline or the call-to-action. Both sit directly in the visitor’s decision path and are common reasons people convert or leave. Test one element at a time so you know which change caused the result.

Keep the original version and treat the test as useful information. You learned the change doesn’t move the needle for your audience. Use that to form your next hypothesis and test something else.

Ready to build?

Put it to work on your own page

Build, test and improve landing pages in one place.