In short
Learning goal: by the end of this lesson you can design a landing page test for one paid traffic source, pick between a split URL and an on-page test, size it on that source's traffic, and report the result to the media buyer without counting the same revenue twice.
Who this is for: agency strategists who work alongside a client's media buyer, in-house or at another agency.
Paid traffic is the easiest traffic to test on and the easiest to misread. It arrives with a promise made by the ad, its volume changes whenever the budget does, and three different tools will each tell you how much revenue it made.
What to measure at this step
| Metric | Definition | Primary goal or diagnostic |
|---|---|---|
| Revenue per visitor (RPV) of everyone who landed | Revenue from visitors who entered the test on the landing page ÷ those visitors | Primary, or profit per visitor where costs are known |
| Next-page rate | Share who view a second page | Diagnostic |
| Product views per visitor | Product pages viewed ÷ visitors | Diagnostic |
| Bounce and scroll | Leave without a second page; how far they scroll | Diagnostic only, never a decision |
| Page errors and load time | Errors and speed per arm | Guardrail |
A lower bounce rate is not a win. A page can keep more people around and sell less. Judge the test on what everyone who landed went on to spend.
Step 1: Match the page to the ad's promise
Each source makes a different promise. A Meta ad showing one product in use promises that product. A Google Shopping click promises a price. A brand search promises the store. The landing page test should ask whether keeping that specific promise sells more.
Work with the media buyer to list the top three campaigns by spend, and for each one write down:
- The promise in the ad (product, offer, problem solved).
- Where it lands today.
- What a shopper must do on that page to buy.
Pick the campaign with the most traffic and the biggest gap between promise and page. That's the first test.
Step 2: Split URL or on-page?
| On-page test | Split URL test | |
|---|---|---|
| What changes | Elements on the existing page | A different page or template entirely |
| Good for | Headline, hero image, offer, order of sections | A new landing page built in a page builder or a different template |
| Risk | Flicker if the page loads slowly | A redirect adds a step; check that query strings, including UTM tags, survive it |
| Measurement | Same URL for both arms | Different URLs; the test assigns the arm before the redirect |
If the media buyer already built a new page, a split URL test is the honest comparison. If you're changing the copy and order of an existing page, test on the page.
Step 3: Target the source, not the whole store
A landing page test for one campaign should include only that campaign's visitors. Target by UTM tag or traffic source so organic and email visitors don't dilute it. Agree with the media buyer that UTM tags stay the same for the length of the test.
Then size the test on that source's traffic alone. Paid traffic often converts below the store average, which makes it slower to test.
(an illustrative client campaign, not a real store)
A Meta campaign sends 1,500 visitors a day to a product landing page, and 2% of them buy. Two arms split the traffic evenly, at 95% confidence and 80% power:
| Smallest change worth detecting (relative) | Visitors per arm | Days |
|---|---|---|
| 15% (2.0% to 2.3%) | 36,693 | 49 days |
| 20% (2.0% to 2.4%) | 21,109 | 29 days |
Seven weeks is longer than most campaigns keep the same creative and budget. A 20% change is a bold redesign, not a headline tweak, but it can be read in about a month. The honest plan for this client is a substantial page change, judged in four to five weeks, with the media buyer agreeing to keep budget and targeting steady for that time.
The figures come from the same sample-size formula Benchmyrk uses to judge results; the sample size calculator runs them for any campaign.
Step 4: Keep the media plan steady
A landing page test assumes both arms get the same kind of visitor. If the campaign's audience changes halfway, both arms still get the same mix, so the comparison stays fair, but the result describes a blend of two audiences. Ask the media buyer to flag any change in budget, audience or creative during the test, and note it in the report.
Step 5: Report without double counting
Three tools will report revenue from the same campaign: the ad platform, the client's analytics, and the testing tool. They count differently. Google Analytics, for example, offers data-driven and last-click attribution models, and the same path can give all the credit to different channels depending on the model (Google Analytics Help, read October 12, 2026). Ad platforms use their own windows.
Rules that keep the numbers honest:
- The test reports a difference between arms, not campaign revenue. "Visitors from this campaign who saw the new page spent more per visitor, by this much, with this 95% interval" is a test result. "The new page made this many dollars" is not.
- Don't add the test's lift to the platform's reported return on ad spend. The platform already credits itself with some of the same orders.
- Report the lift for the campaign's traffic only. It doesn't apply to the rest of the store.
- Send test exposures to analytics so the media buyer can see the arms in their own reports, labelled as test arms.
Common problems
- Judging on bounce rate or time on page. Diagnostics, not outcomes.
- Mixing sources. A test on all traffic hides the paid effect inside organic visitors.
- Changing the campaign mid-test. Budget and audience changes make the result describe a blend.
- Losing UTM tags in a redirect. Then targeting and reporting both break.
- Double counting. The test's lift is not extra revenue on top of the platform's numbers.
Test ideas, with the mechanism behind each
| Idea | Mechanism | Judge on |
|---|---|---|
| Headline that repeats the ad's promise | Confirms the shopper is in the right place | RPV of everyone who landed |
| The advertised product first, not the collection | Removes a search step | RPV; product views as a diagnostic |
| The offer in the ad shown above the fold | Keeps the price promise | RPV; check the offer's cost with profit per visitor |
| A dedicated landing page against the product page | Tests the whole page, via split URL | RPV |
A 10-minute exercise on your own client
- Ask the media buyer for the top campaign by spend, its daily visitors and its conversion rate.
- Open the ad and the landing page side by side. Write down the promise and whether the page keeps it.
- Put the campaign's traffic and conversion rate into the sample size calculator at 15% and 20%.
- Decide: is this a bold page change you can read in a month, or a fix to make without a test?
Do it in Benchmyrk
Testing landing pages in November? Part 9 of the agency BFCM series covers what not to test for clients over BFCM; find it on the BFCM hub.
Check your understanding
- The new page lowered bounce rate by a third and RPV is flat. Did it win? No. The primary goal is revenue per visitor of everyone who landed. Bounce rate is a diagnostic.
- The media buyer doubled the budget in week 2. Is the test broken? Not broken: both arms still get the same mix. But the result now describes a blend of audiences, so note it in the report.
- The test shows a significant rise in RPV. Can the media buyer add that percentage to the campaign's reported revenue? No. The platform already credits itself with some of those orders. Report the lift as a difference between arms, with its interval.