Hypothesis
If reviews or testimonials sit beside the primary call to action instead of lower on the page, more visitors will convert, because the proof meets the doubt at the moment of decision. The decision is whether to place reviews, ratings or testimonials beside the call to action, or leave them lower on the page, as a standard for a page template.
The reasoning comes from how proof works. A visitor's doubt peaks when they are about to commit: the button, the form, the price. The trust and proof elements guide sets out which kinds of proof answer which doubts and why placement near decisions matters, and usability research such as the articles published by Nielsen Norman Group describes people scanning a page rather than reading it, so proof several screens away from the button may never be seen at the moment it would help. The test checks whether that reasoning holds on your page, with your visitors. Write the page, the proof element and its two positions into the experiment hypothesis template before building anything.
Metrics and guardrails
The primary metric is the conversion rate of the page's primary action: purchases, form submissions or sign-ups divided by visitors assigned to each arm. Count visitors, not sessions, so that a returning visitor is counted once in the arm they were assigned to.
Three guardrails catch the ways a placement can win on paper and lose in practice. Average order value, for stores, or the lead qualification rate, for lead generation, checks that the extra conversions are not smaller or weaker ones; proof next to the CTAThe prompt that tells a reader the one action to take next, such as a button or link. can nudge hesitant visitors who then buy less or turn out to be poor leads. Page load time per arm checks the build: review widgets often load scripts and shift the layout, and a variant that loads the widget earlier or higher on the page can be slower than the control. Clicks through to the full reviews show whether the moved proof is being used; a rise in conversion with no change in review clicks suggests the summary did the work, while a fall in review clicks with no gain in conversion suggests visitors saw enough without reading further.
Design
Randomise at the level of the visitor, assign on first view and keep each visitor in the same arm on return visits. The control keeps the current layout. The variant moves the existing proof element, or a compact form of it such as the rating and review count already shown elsewhere, so that it sits beside the primary call to action. Nothing else changes: same headline, same copy, same images, same proof content, same offer.
| Design element | Choice | Reason |
|---|---|---|
| Unit of randomisation | Visitor, assigned on first view and kept in one arm | Visitors return before converting |
| Arms | Current layout, against proof moved beside the call to action | One placement change |
| Proof content | Identical in both arms; moved, not added or rewritten | Isolates placement from content |
| Held constant | Headline, copy, images, offer, form, load order of scripts | Keeps the comparison clean |
| Devices | One placement rule per device, written before launch | Mobile and desktop layouts differ |
| Primary metric | Conversion rate of the primary action | The decision is about conversions |
Choose a template rather than a single page where possible. On a store, the product page optimisation guide describes what belongs near the button on a product page, and a product template carries enough traffic to test where a single Landing PageThe page a visitor arrives on from a campaign, built to turn that visit into one action. may not. Baymard Institute's e-commerce research and the guide to reviews and social proof cover how shoppers use ratings and reviews on product pages; this test borrows the display and changes only where it sits.
Sample size reasoning
A placement change usually has a modest effect, so the minimum detectable effect will be small and the sample large. The inputs are the page's current conversion rate, the smallest rise that would justify changing the template, and the confidence the team requires. The sample size and significance guide explains how those inputs combine and why a small effect on a low baseline needs many visitors, and the conversion rate calculator set up for landing page tests shows the rates the plan starts from. If the page's traffic cannot carry the test in a reasonable time, test across a whole template, or make the decision on design grounds and skip the test.
Duration
Run at least two full weekly cycles and until the planned sample is reached, then read once at the end. Whole weeks matter because weekday and weekend visitors arrive with different intent and devices. Fix the sample and the end date before launch, and avoid a sale or campaign that changes the page's traffic mix midway.
Reading the results and the decision rule
Read once the planned sample is reached. Compare conversion rates with their confidence interval, then check the guardrails, then the device split named before launch. The A/B testing guide covers the wider reading discipline. The rule, written before launch:
| Result | Decision |
|---|---|
| Variant converts more and guardrails hold | Ship the placement across the template |
| Variant converts more but order value or lead quality falls | Keep the current layout, or re-test with the proof beside the price or form only |
| Variant is slower to load | Fix the widget's loading and re-test before deciding |
| Result inside the noise | Keep whichever layout the design system prefers |
Once the test has finished, the significance checker below reads the confidence of the difference between the two conversion rates from each arm's visitors and conversions. It reads a finished two-arm test; it does not plan one or tell you how many visitors you need.
A/B test significance checker
Pitfalls
Changing the proof as well as its position is the most common way to spoil the test. A designer moving the reviews often picks the best testimonial, shortens the summary or adds a badge, and the variant then differs in content as well as place. Lock the proof content before the build and compare the two arms side by side before launch.
Layout shift from review widgets is the second. Many review tools inject their content after the page has loaded, so a widget moved higher pushes the button down as it appears, causing mis-taps and a slower-feeling page. Reserve the space for the widget and check the guardrail on load time. The third is mobile and desktop placements diverging. A layout that puts proof beside the button on desktop may place it above or below the button on a phone, so the arms test different things on different devices. Write one placement rule per device into the plan and check both on real phones before launch.
A mattress retailer moves its rating beside the button
Ochreworth Mattresses, an invented online mattress retailer, shows its review summary near the bottom of each product page. It tests moving the existing star rating, review count and the three most recent reviews beside the add-to-basket button, with no other change. Over four weeks, 46,000 visitors are split evenly. The control converts at 2.10% and the variant at 2.42%, with a confidence of about 98% from the significance checker. Average order value is unchanged, load time is within 50 milliseconds between arms after the team reserves the widget's space, and clicks through to the full reviews rise. Ochreworth ships the placement on every product page.
Common mistakes
Placement tests fail when they change more than the placement. The landing page headline test changes the opening line instead, and the experiment library lists every design. A placement win says where Social ProofEvidence that other people have chosen, used or approved something, used to reduce a buyer's doubt. works on this template, not which proof works best; that is a separate test.
- Rewriting the proof while moving it. The test then measures content and position together.
- Ignoring load time. A slower variant can lose for reasons that have nothing to do with proof.
- Reading devices that were not planned. Decide the device split before launch.
- Stopping on a good day. Read once, at the planned sample.
- Testing on a page too quiet to answer. Use a template, or decide on design grounds.
- Treating the significance checker as a planner. It reads finished tests only.
