A landing page A/B test that looks for a 20% relative lift on a 5% conversion rate needs about 7,663 visitors per variant, or roughly 15,300 paid clicks for a two-way test. At the 2026 average search cost per click of $5.42 reported by LocaliQ and WordStream, that is about $83,000 of media before anyone can call a winner. Most agency proposals promise "ongoing testing" without doing that sum.
This guide is for paid-media teams scoping a landing page design agency. It covers the four things the agency should hand over (copy, build, tracking and a test plan) and how to check each before launch, using Google's and Meta's own documentation, the testing vendors' published data and our own sample-size calculation.
The four deliverables, and what done looks like for each
A paid-media team does not buy design. It buys a page that turns clicks into qualified pipeline and proves it did. That takes four outputs, each checkable before launch.
| Deliverable | What you receive | How to check it | The usual gap |
|---|---|---|---|
| Copy | A copy document mapping each headline and offer to an ad group | Read each ad next to its page | One generic page for every ad group |
| Build | A live page on your domain, built from reusable sections | Good Core Web Vitals on mobile; no crawl errors | Staging password left on, or only the agency can edit |
| Tracking | A tracking map of every event and the account that owns it | A test lead appears in Google Ads, Meta and the CRM | Conversions fire on a page view and nothing reaches the CRM |
| Test plan | Metric, baseline, minimum lift, visitors per variant, duration, stop rule | Recompute the sample size from its inputs | "We will test headlines" with no traffic estimate |
Source: Google Ads Help and Meta for Developers pages listed below; checklist compiled by Octo Agency
If a proposal does not name all four, ask which ones it expects your team to cover. Many paid-media teams already own tracking and testing and only need copy and build. That split is fine, provided it is written down.
Copy: the ad's promise, repeated above the fold
Google's advice to advertisers is to choose a page that closely matches the ad and keywords, and to mirror the ad's call to action on the page. That is message match, and it is a copy deliverable before it is a design one.
Map copy to ad groups. The agency should receive your ad groups, top search terms and ad copy before it writes anything. The output is one row per ad group or audience: headline, subhead, offer, call to action and proof point. Three ad groups can share a page through dynamic text or variants. They should not share a brand headline.
Write plainly. Unbounce, a landing page builder, analysed more than 41,000 pages for its 2024 Conversion Benchmark Report. Pages at a 5th to 7th grade reading level converted at 11.1%, 56% higher than pages at an 8th to 9th grade level and more than twice the 5.3% for professional-level writing. Word count was negatively correlated with conversion. Financial services was the one industry where more complex language could help.
Respect the intent. Unbounce's average conversion rates were 10.9% for paid search and 12% for paid social, against an all-industry median of 6.6%. Paid visitors expect a specific answer; a page that answers a different question wastes that intent.
Say who writes it. The scope should state whether the agency writes copy or lays out yours, how many revision rounds are included, and who on your side approves claims, logos and numbers.
Build: a page Google can crawl and a phone can load
It must pass Google's destination requirements. Google disapproves ads whose destination returns an error to its AdsBot crawler, cannot be crawled, or does not match the display URL's domain. A staging password, a redirect to another domain or a tracking template that lands on different content will all trip this. Launch on your own domain and confirm AdsBot can fetch the page before the campaign starts.
It must be fast on a phone. Google's Core Web Vitals call a page good when Largest Contentful Paint is within 2.5 seconds, Interaction to Next Paint within 200 milliseconds and Cumulative Layout Shift 0.1 or less, at the 75th percentile of page loads. Portent's 2022 study of 14 B2B lead generation sites found a page loading in 1 second converted at three times the rate of one loading in 5 seconds. Fourteen sites is a small sample, so treat it as a direction.
Quality Score is a symptom. Landing page experience is one of three Quality Score components, rated above average, average or below average. Google states that Quality Score is not an input in the ad auction. A poor rating is a reason to reread the page against the ad, not a target for the contract.
Your team must be able to edit it. Paid-media teams launch pages faster than agencies build them. Ask for reusable sections and a short guide to combining them, so a variant for a new ad group is an afternoon in the editor.
State the platform. Landing page builders, Webflow, WordPress and custom code all work. We build marketing sites and landing pages in Webflow and applications in Next.js, so read this with that bias: what matters is whether your team can publish without the agency.
Tracking: the part most scopes leave vague
Paid-media teams are judged on cost per qualified lead, and the landing page is where the click ID is captured or lost. "Set up tracking" gets read as "install the tags". Name the parts.
| Component | What it does | Verify by |
|---|---|---|
| Consent mode | Google tags follow the visitor's choices on ad_storage, analytics_storage, ad_user_data and ad_personalization | Check tag behaviour before and after the banner choice |
| Enhanced conversions for leads | Sends a SHA256-hashed, normalised email from the form | Google Ads diagnostics show hashed data arriving |
| GCLID capture | Stores the Google Click ID in a hidden field and the CRM record | A test lead shows the GCLID in the CRM, case preserved |
| Offline conversion import | Sends qualified and won stages back against the stored GCLID | Imported conversions appear in the account |
| Meta Pixel plus Conversions API | Sends each lead from browser and server with a shared event_id | Events Manager shows one lead, not two |
Source: Google Ads Help, Google for Developers and Meta for Developers pages listed below
Google offers two consent mode implementations. In basic, tags wait until the visitor uses the banner. In advanced, tags load with consent denied by default and send cookieless measurements until consent is granted, which supports a more detailed advertiser-specific model. Which one fits is a legal decision for your side.
Google's GCLID import requires auto-tagging, a GCLID stored with each prospect, and a click-to-conversion cycle under 14 or 90 days, depending on the data source. A deal that closes in five months will not import against its click, which is why the qualified stage usually becomes the optimisation target. On Meta, the recommended setup sends each event from both the Pixel and the Conversions API, and Meta deduplicates only when event name and event ID match within 48 hours.
Our guide The Analytics Stack for a Four-Week Site Build covers GA4, server-side tagging and dashboards. For a landing page scope, the tracking map is the deliverable to insist on.
The test plan: what your clicks can actually pay for
A test plan should state the primary metric, baseline rate, smallest lift worth detecting, visitors per variant, expected duration and stop rule. With those six numbers, anyone can check it.
We calculated the sample sizes below with the formula behind Evan Miller's sample size calculator, at its default 80% power and 5% significance, and priced clicks at LocaliQ and WordStream's 2026 all-industry search average of $5.42.
| Baseline rate | Lift to detect | Visitors per variant | Total clicks | Cost at $5.42 |
|---|---|---|---|---|
| 3% | 20% relative | 13,050 | 26,100 | $141,464 |
| 5% | 10% relative | 30,244 | 60,489 | $327,849 |
| 5% | 20% relative | 7,663 | 15,326 | $83,067 |
| 5% | 30% relative | 3,450 | 6,900 | $37,397 |
| 8.18% | 30% relative | 2,034 | 4,068 | $22,050 |
Source: Octo Agency calculation using Evan Miller's sample size formula and LocaliQ and WordStream 2026 search benchmarks, read 2026-10-21
LocaliQ's figures come from its own customers' Google and Microsoft Ads campaigns, so they are an ad agency's data, not a census. Its 8.18% conversion rate counts every Google Ads conversion type, which is why that row is the optimistic one. Business services averaged 4.85% at $5.87 a click, close to the 5% rows.
Small lifts are normal. Optimizely, which sells experimentation software, reports that 12% of 173,000 experiments produced a significant improvement on the primary metric, with losses about as common as wins. Convert, another testing vendor, reports that 60% of completed tests show under 20% lift. Most landing page tests on search traffic cannot reach a verdict on a typical budget.
What a realistic plan does. It tests large changes, such as a new offer, a shorter form or a new structure, where a 30% swing is plausible. It tests on cheaper traffic first and carries winners to search. It can use Google Ads custom experiments, which Google says are typically used to test landing pages, so the split happens at the campaign. And it says in advance what happens when a test is inconclusive.
Recompute before you sign. A promise of a test every fortnight on 2,000 clicks a month means very large changes or winners called early. Our A/B Testing Statistics 2026 piece covers the wider win-rate data.
What to write into the statement of work
These are the lines we would expect in a landing page scope, and the ones teams most often find missing later.
- Pages and variants. How many, and whether a variant means new copy, a new layout or a new offer.
- Turnaround. Working days from brief to live for a new ad group's page once the system exists.
- Revisions and approvers. Rounds per deliverable and one approver on each side.
- Ownership. Pages on your domain, in an account you own; copy, design files and the tracking map transfer at the end.
- Acceptance. The test lead above, run and recorded before launch.
- After launch. Whether new variants and test analysis sit in a retainer, are priced per request, or are out of scope.
For prices and testing evidence, our teardown 15 Landing Page Agencies Checked on Their Own Pages compares agencies on their own sites. Most quote per project, and so do we.
What this means for a paid-media team hiring a landing page agency
Start from your traffic, not the agency's portfolio. Pull clicks, conversion rate and cost per click for the campaigns the pages will serve, and work out what a test can detect before anyone promises one. Then scope the four deliverables by name, with the tracking acceptance test as the condition for launch.
We build marketing sites and landing pages in Webflow, run a four-week process for a marketing site, and have delivered more than 20 Webflow sites in twelve months as a white-label partner. On a landing page brief we would hand over the copy map, the section library, the tracking map and a test plan with its sample size written out, and flag a test your traffic cannot fund before the campaign starts. Our marketing site and landing page service page explains how we scope one.
