---
title: Testing landing pages on Meta — dedicated ads per page instead of A/B split tools
slug: landing-page-tests-with-dedicated-ads
summary: Barry Hott's method for testing big landing-page changes with optimized Meta traffic — duplicate your top relevant evergreen ads (same creative, copy, targeting) into new ad sets pointing at the new page, let Meta optimise each combo, compare to business as usual — and why external 50/50 split tools dilute relevance signals, lag for 2+ weeks and falsely penalise better pages; plus site pages vs landing pages and when small A/B tests are still fine.
domain: marketing
tags: [landing-pages, ab-testing, meta-ads, cro, relevance, testing]
related: [ad-relevance-chain, judging-creatives-by-spend-not-roas, meta-landing-page-congruency, ad-landing-page-message-match, persona-landing-pages-and-listicles, ads-get-clicks-no-conversions, meta-ads-per-ad-set-and-budget-minimums]
sources: [yt:oFZp553oL2c]
updated: 2026-10-24
---

# Testing landing pages with dedicated ads

## TL;DR
- **Meta doesn't know you're split-testing your destination.** It routes traffic using past ad + page + user signals. Sending 50% of that traffic to a different page breaks the relevance it learned.[^1]
- With **optimized** traffic (a conversion goal), the page changes who converts, and so who Meta targets next. With unoptimized (traffic) campaigns, the destination barely affects delivery.[^2][^3]
- For **big swings** (new page, new storytelling, new visuals, a headline that changes the audience), duplicate your top relevant evergreen ads into new ad sets, point them at the new page, and let Meta optimise that combination.[^4][^5]
- Don't run it as a Meta A/B test either. Benchmark against business as usual (CPA, spend at a cost cap, ATC/IC rates).[^6]
- Sometimes **A and B live together beat A or B** — each page finds its own audience.[^7]

## Details
**Site pages vs landing pages.**[^8][^9][^10]
| | Site pages (home, category, PDP) | Landing pages |
|---|---|---|
| Built for | every source and life-cycle stage | ad traffic, can target cold or warm |
| Stakeholders | many | few (CRO team, media buyer, agency) |
| Data | mixed with direct/organic/SEO, which inflates CVR | clean — one or a few sources |
| Navigation | full site nav, logo, cart = exits | optional; can force one guided flow (e.g. Rotten's fake nav only scrolls to the offer) |
| Changes | affect everyone | affect only relevant traffic; editable without devs (Replo, PageDeck-type builders) |

More cold traffic lowers conversion rate without the page being worse. Don't judge ad pages against site averages.[^11]

**Changes that shift relevance** (so they need a dedicated-ad test): headline "for busy moms" → "for elite athletes"; cozy-home imagery → intense gym; hero model age, gender, skin colour, number of people; and above all a CTA change from *Learn more* (scroll) to *Buy now* (straight to checkout), which suits warm visitors and loses cold ones.[^12][^13]

**What goes wrong with split tools.** They dilute relevance signals and make the whole account volatile, and the testing tool alone can slow delivery, so run an A/A test first.[^14] There is **relevance lag**: two weeks often isn't enough for the system to adjust. The tools can also *falsely penalise* the better page. In one of Barry's tests, a research-backed new page (heat maps, customer interviews) lost the split test and hurt overall performance. Re-run as separate ads, it performed "amazing" with a different audience.[^15][^16]

**Procedure.**[^4][^5][^17]
1. Pick the top-spending evergreen ads that are *relevant to the page's topic* (not just the top spenders).
2. Duplicate with identical creative, copy, targeting and bidding into new ad sets/campaigns; change only the URL.
3. Target broad so Meta can find the sub-audience that the new page suits.
4. Watch spend, CPA and secondary metrics (ATC, initiate checkout). If the new page makes ATC easier, those rates aren't comparable one-to-one.
5. If your best ad's social proof can't be duplicated, rebuild it twice (old page vs new page) so both start equal.
6. Re-test past "losing" pages this way.[^18]

**When not to.** With limited conversion data (high AOV, low spend) the extra ads fragment the account, so stay consolidated. Small changes that don't shift relevance are fine in a normal external A/B tool. Ask whether the headline you tested changed *who* the page is for.[^19][^20]

## Disagreements / caveats
- A single anecdote supports the "falsely penalised page" claim, and Barry offers no numbers. The method also confounds page and audience on purpose: the goal is "scale, not science".[^7]
- Warm-audience bias: the page with the best CVR may just be favoured by warm traffic. Barry says to build for both cold and warm.[^19]
- Ad-landing congruency notes argue for one page per angle ([[meta-landing-page-congruency]]); this note is about *how to test* those pages.

## See also
- [[ad-relevance-chain]] · [[meta-landing-page-congruency]] · [[persona-landing-pages-and-listicles]]
- [[people/barry-hott]]

[^1]: yt:oFZp553oL2c @ 26:43 — Barry Hott
[^2]: yt:oFZp553oL2c @ 12:53 — Barry Hott
[^3]: yt:oFZp553oL2c @ 16:22 — Barry Hott
[^4]: yt:oFZp553oL2c @ 36:20 — Barry Hott
[^5]: yt:oFZp553oL2c @ 37:07 — Barry Hott
[^6]: yt:oFZp553oL2c @ 41:55 — Barry Hott
[^7]: yt:oFZp553oL2c @ 40:16 — Barry Hott
[^8]: yt:oFZp553oL2c @ 00:46 — Barry Hott
[^9]: yt:oFZp553oL2c @ 08:05 — Barry Hott
[^10]: yt:oFZp553oL2c @ 12:07 — Barry Hott
[^11]: yt:oFZp553oL2c @ 06:25 — Barry Hott
[^12]: yt:oFZp553oL2c @ 18:38 — Barry Hott
[^13]: yt:oFZp553oL2c @ 20:56 — Barry Hott
[^14]: yt:oFZp553oL2c @ 29:57 — Barry Hott
[^15]: yt:oFZp553oL2c @ 30:44 — Barry Hott
[^16]: yt:oFZp553oL2c @ 34:44 — Barry Hott
[^17]: yt:oFZp553oL2c @ 38:38 — Barry Hott
[^18]: yt:oFZp553oL2c @ 45:00 — Barry Hott
[^19]: yt:oFZp553oL2c @ 41:02 — Barry Hott
[^20]: yt:oFZp553oL2c @ 42:44 — Barry Hott