Most teams add testimonials to a landing page, watch conversions tick up, and assume the testimonials caused it. Sometimes they did. Often the page changed in five ways at once, traffic quality shifted, or the sample was too small to mean anything. If you want to know whether a testimonial actually earns its place — and which testimonial earns the most — you have to test it properly. This guide walks through how to run an A/B test on the testimonials in your landing page so the result is real and not a story you told yourself.

What is actually worth testing

Not every testimonial change is worth an experiment. A/B testing has a cost — traffic, time, and the risk of shipping a worse variant to half your visitors — so reserve it for decisions that recur and matter. The changes that repay a test are:

Presence vs. absence — does a testimonial block near the CTA lift conversion at all, versus no social proof there?
Placement — the same testimonial above the fold vs. beside the pricing table vs. under the CTA.
Which testimonial — a results-driven quote vs. a trust-driven quote in the same slot.
Format — a plain text quote vs. the same quote with a face, name, and company logo.
Length — a one-line pull quote vs. a three-sentence story.

Everything else — font tweaks, punctuation, a slightly different crop — is not worth the traffic it would burn. Test decisions you will reuse, not one-off cosmetics.

The one-variable rule

The single most common way teams fool themselves is changing more than one thing between variant A and variant B. If B has a new testimonial and a new headline and a repositioned CTA, and B wins, you have learned nothing about the testimonial. You have learned only that the bundle beat the old bundle. A valid test isolates exactly one variable: the two variants must be identical in every respect except the thing you are measuring. If you want to test both a new testimonial and a new headline, that is two sequential tests, not one. This discipline feels slow, and it is the only thing that makes the result trustworthy.

Pick one metric before you start

Decide your success metric before the test runs, and write it down. The right metric is almost always the action the page exists to drive — a signup, a demo request, a purchase — not a proxy like time on page or scroll depth. Proxies are dangerous because a variant can improve them while hurting the thing you actually care about; a longer, more engaging testimonial might raise scroll depth and lower signups because it delays the CTA. Choose the downstream conversion, commit to it in advance, and ignore the proxies. Deciding the metric after you see the data is how a losing variant gets rescued by a metric that happened to move.

How much traffic you need

The hardest truth about A/B testing testimonials is that small sites often cannot run a valid test at all, because the effect is too small to detect without a lot of traffic. A testimonial change might move conversion by one or two percentage points — a real, valuable lift — but detecting a lift that size with confidence takes thousands of conversions per variant, not dozens. Before you start, estimate: if your page converts a few hundred visitors a week, a test that needs 20,000 visitors per arm will run for months, during which everything else about your business changes and pollutes the result.

The practical rules:

Use a sample-size calculator with your current conversion rate and the smallest lift worth caring about, and get the required number before launching.
Run for whole weeks, never partial ones, so weekday and weekend traffic are represented equally in both arms.
Do not peek and stop early. Calling a winner the moment it looks ahead is the classic way to ship noise as a result — early leads reverse constantly.
If you cannot reach the sample size in a reasonable window, do not A/B test. Use judgment and best practices instead, and be honest that you are choosing, not proving.

The mistakes that produce fake wins

Four errors turn A/B tests into confident nonsense:

Stopping when you like the number. Fixing the sample size and end date in advance is the entire defense against this.
Testing during an anomaly. A launch, a press mention, or a holiday skews traffic quality; a testimonial that "wins" during a spike may lose in normal conditions.
Ignoring segments. A testimonial from an enterprise customer might lift enterprise visitors and depress small-business ones, netting to zero. If the aggregate is flat, look at the segments before concluding "no effect."
Running ten tests and celebrating the one that won. Run enough tests and something wins by chance. If you test many testimonials at once, expect a few false positives and re-run the apparent winner to confirm.

After the test: keep or kill, then document

When the test reaches its predetermined sample size, read the result honestly against the metric you committed to. If the new testimonial won, ship it and note why it likely won — was it the specificity, the named customer, the placement? That note compounds into a house style over time. If it lost or came out flat, that is also a result: revert, and record that this type of testimonial did not move this page. A losing test is not a failure; it saved you from shipping a worse page on a hunch.

How to A/B Test Testimonials on a Landing Page (Without Fooling Yourself)