A/B testing landing pages properly means changing one meaningful thing, splitting comparable traffic evenly between versions in your tracker, deciding in advance how many conversions you need, and only calling a winner once the test reaches that sample and a full weekly cycle. For affiliate campaigns, judge the result on revenue per visitor (or EPC), not on lander click-through rate alone, because the version that gets more clicks does not always earn more.

Most failed tests do not fail on the idea. They fail because they were stopped too early, run on mixed traffic or measured on the wrong number. This method avoids all three.

What to test on affiliate landers and offers

Test things big enough to change behaviour. Button colour tests on small traffic rarely produce a result you can trust. Good candidates, roughly in order of likely impact:

  1. The offer itself: two programs or two payout models behind the same lander.
  2. The angle: the promise or reason to click, such as price-led versus feature-led copy.
  3. Lander versus direct linking: does a pre-sell page help, or just add a step?
  4. Page structure: a quiz or step-by-step lander against a single-page layout.
  5. Headline and main call to action.
  6. Trust and clarity elements: pricing clarity, plain-language terms and clear disclosure.

For ideas specific to adult offers, see our guide to adult landing pages that convert. Whatever you test, keep every version honest: a lander that wins by misleading visitors about price or content creates refunds, chargebacks and regulatory risk that the test will not show.

Step 1: Write a hypothesis

A hypothesis forces you to say what you expect and why, which makes the result useful even when the test loses. Use this format:

Because [observation], we believe that [change] will increase [metric] for [audience]. We will know when [sample and threshold].

Hypothetical example: "Because mobile visitors scroll past the comparison table, we believe moving the top recommendation above the fold will increase revenue per visitor for UK mobile traffic. We will know after 400 conversions per variant."

Step 2: Split test landing pages in your tracker

Trackers are built for this. In most, you add several landers or offers to one campaign path and assign weights.

  • Use equal weights (50/50, or 33/33/33 for three versions). Uneven splits slow the test down.
  • Keep everything else fixed: same traffic source, GEOs, devices, bids and placements. Changing bids mid-test changes the audience.
  • Split by visitor, not by click, where your tracker allows, so a returning visitor sees the same version.
  • Pass the variant into the offer link as a sub-ID, so network conversions can be matched to the lander. See sub-IDs explained.
  • Test postbacks for every variant before you spend money. A broken postback on one version produces a false loser.

For organic content pages, you can run tests with a dedicated testing tool or your CMS. Google Optimize, once a common free option, was shut down in September 2023, so older tutorials that rely on it are out of date. If you are still choosing a tracker with good split-testing tools, read our comparison of the best affiliate trackers.

Step 3: Choose the right metric

MetricUse whenWatch out for
Lander click-through rateEarly signal, high volumeMore clicks can mean lower-quality clicks
Conversion rate (visits to conversions)Same offer and payout in every variantIgnores payout differences
Revenue per visitor or EPCComparing offers, payouts or anglesNeeds enough conversions; noisier
ROIPaid traffic, final decisionDepends on cost data being accurate

Pick one primary metric before the test starts and stick to it. If you are testing offers with different payouts, revenue per visitor is the only fair comparison. The EPC maths is explained in our guide to calculating EPC.

Step 4: Work out the landing page test sample size

Decide how much data you need before you start. A widely used rule of thumb, for a test with 80% power at the 5% significance level, is:

Visitors needed per variant ≈ 16 × p × (1 − p) ÷ d²

where p is your current conversion rate and d is the smallest absolute improvement you care about detecting.

Hypothetical example: your lander converts at 5% (p = 0.05) and you want to detect a lift to 6% (d = 0.01). That gives 16 × 0.05 × 0.95 ÷ 0.0001 = 7,600 visitors per variant, or roughly 380 conversions each. To detect a lift from 5% to 5.5%, you would need four times as many visitors.

The lessons are simple. Small improvements need big samples. Low-traffic sites should test bold changes, not small tweaks. An online sample size calculator will give more precise figures, but the rule of thumb tells you quickly whether a test is realistic.

Step 5: Check statistical significance properly

Statistical significance tells you how unlikely your result would be if the two versions actually performed the same. A common threshold is 95% confidence (a p-value below 0.05). Many trackers and free online calculators run a two-proportion test for you.

Three rules matter more than the exact threshold:

  1. Do not peek and stop. Checking every day and stopping the moment you see 95% inflates false winners dramatically. Decide the sample in advance and judge the result when you reach it, or use a tool built for sequential testing.
  2. Look at the size of the effect. A significant 0.1% lift may not be worth the effort of maintaining a variant.
  3. Watch for revenue noise. A few large RevShare customers can swing revenue per visitor. Check that a winner is not driven by two or three outlier sales.

When to stop a test

  • Stop and declare a winner once you reach the planned sample, the test has run for at least one full week (two is better, to cover weekday and weekend behaviour) and the result is significant on your primary metric.
  • Stop early for harm if one variant is clearly broken, for example zero conversions after the point where the other has dozens. Check tracking first.
  • Stop and call it a draw if you reach the sample with no significant difference. Keep the simpler version and test something bolder.
  • Restart if anything major changed mid-test: offer payout, traffic source mix, tracking or a site outage.

Re-test winners occasionally. Traffic sources, seasons and audiences change. A lander that won in spring may fade by autumn, so give your current champion a fresh challenger every few months.

A/B testing offers, not just pages

Offer tests follow the same method with one difference: you usually cannot see the full value of a conversion straight away. A CPA offer pays now; a RevShare offer pays over months. Compare offers on revenue per visitor over the same time window, and revisit RevShare results after 60 or 90 days. Rotations and smartlinks automate some of this, but they hide the detail of why one offer wins.

Frequently asked questions

How long should an affiliate landing page test run?

Until it reaches the sample size you calculated, and for at least one full week so weekday and weekend visitors are both included. Two weeks is safer. If your traffic cannot reach the sample within about a month, test a bolder change instead.

Can I test more than two versions at once?

Yes, but each extra variant needs its own full sample, so a three-way test needs roughly 50% more traffic than an A/B test. Comparing many variants also raises the chance of a false winner, so keep multi-variant tests for high-traffic campaigns.

Should I trust my tracker's winner label?

Treat it as a starting point. Check which metric and confidence level it uses, whether the sample is big enough and whether the result holds for each main GEO and device. Automatic optimisation that shifts traffic to the leader can also bias the comparison.

Next steps

Write down your next hypothesis, calculate the sample and check whether your traffic can reach it within a month. Record every result, winners and losers, in your reporting dashboard. To compare trackers with strong split-testing features, see our affiliate tracking platforms directory.