# Landing page design best practices: what the evidence actually supports

> A landing page should match the intent behind an ad group, not one ad, because responsive search ads rotate their headlines. Most other rules, such as removing navigation, rest on vendor tests with no reported significance. With 100 visitors a day, a 5 percent page cannot reveal anything under a 60 percent lift in three weeks.

By LAIN, 16 Mar 2026. 9 min read. Category: Website. Canonical: https://lainagency.com/resources/landing-page-design-best-practices

Landing page advice is mostly repeated rather than tested: one goal, no menu, short form, big button. Some of it has a mechanism behind it. Some rests on a vendor's own test with no duration or significance reported. And some of it you cannot verify on your own traffic at all. This guide separates the three, so you spend design time on the parts that are mechanical and treat the rest as hypotheses.

## Match the intent, not one ad
"Message match" usually means the headline repeats the ad. That works only if there is one fixed ad. A Google responsive search ad holds up to 15 headlines of 30 characters and 4 descriptions, and Google assembles them in varying combinations and orders. A visitor may have seen any of them, so the page cannot echo one of them. What it can match is the **promise every combination makes**: the offer and the intent of the search.

| Ad group intent | Headlines that may rotate (30 characters or fewer) | One page promise |
|---|---|---|
| Free website audit | "Book a Free Website Audit" (25), "Website Design for Startups" (27) | A free audit, what it covers, a booking form |
| Brand identity | "Logo and Brand Identity" (23), "Brand Identity in 4 to 6 Weeks" (30) | The identity process, with proof of work |
| Build in Webflow | "Design and Build in Webflow" (27) | What is built, in which tool, how it ships |

Three consequences follow. One page per **intent cluster**, not per ad. Pin only the headline that carries the offer, because Google does not recommend pinning for most advertisers and says it limits the combinations the system can test. And keep one phrase for the offer on the page, in the same words, everywhere it appears.

Quality Score deserves a correction here. Google calls it "a diagnostic tool" and says it is "not an input in the ad auction". Its three parts are expected click-through rate, ad relevance and landing page experience, each rated against other advertisers whose ads showed for the same search in the last 90 days. The auction itself uses auction-time measurements of those same things. So treat Quality Score as a symptom report, not a target.

![An ad card and a page card side by side carrying the same headline, Book a free website audit.](https://lainagency.com/images/blog-art/landing-page-design-best-practices-fig.webp)

## Design the first screen, and test it cold
Nielsen Norman Group's 2018 eye-tracking study (120 participants, more than 130,000 fixations, 1920 by 1080 screens) found people spent about 57% of viewing time above the fold, 74% in the first two screenfuls and 81% in the first three. In 2010 the above-the-fold share was 80%, so people scroll more now, but they still look mostly at the top. NN/g warns that very long pages risk losing attention and recommends testing with real users to find the right length.

Use a test, not a rule. Cover everything below the first screen at 390 by 844 pixels and show it to five people for five seconds. Ask three questions: what is this, who is it for, what would you press. If fewer than four of five answer all three, the headline and first visual are the problem. The threshold is a suggestion, not a published rule.

The first screen has a second owner: the **largest contentful paint**. The web.dev definition counts the largest image, text block or video in the viewport, keeps the last candidate until the visitor interacts, and calls 2.5 seconds or less good at the 75th percentile. A consent banner can be that element. DebugBear measured an LCP rising from 1.43 to 3.61 seconds when the banner's text block became the largest element, in one example. Check which element your trace calls the LCP with the banner present, because visitors who have not yet chosen will see it.

## One action, and the navigation question
Give the page one primary action, labelled with what happens: "Book a 30 minute call" beats "Submit". The evidence for removing navigation is thinner than it looks. HubSpot tested five of its busiest landing pages with all external links removed, including the top navigation, footer navigation and share links. Conversion rose 0 to 4% on the three top-of-funnel offers and 16% and 28% on the two mid-funnel pages. The write-up gives no duration, sample size or significance. Vendor case studies elsewhere report larger gains from single tests. Read all of this as consistent with "removing exits does not hurt", not as a measured size of effect. A logo-only header is cheap and reasonable. Whether to keep legal footer links is our advice, not a tested result: they take little attention, and HubSpot's test removed them too.

Two details are routinely got wrong.

- **Button contrast.** WCAG 2.2 criterion 1.4.11 requires 3:1 for visual information needed to identify a component. W3C says a text-labelled button needs no visible boundary at all, so a pale button on a pale ground is not a failure of 1.4.11 as long as its text identifies it, and the text still has to meet its own text contrast. A boundary needs 3:1 only when it is the only cue. The focus indicator must contrast with adjacent colours either way.
- **Pop-ups and exit overlays.** Google's destination policy lists pop-ups or interstitials that interfere with seeing the requested content, interfering with the back button, and slow loading as non-compliant. Interstitials are acceptable if they do not make it hard to leave. An exit-intent overlay added by a testing tool is where this goes wrong.

## Speed: what the famous study measured
The line "0.1 second faster, 8% more conversions" is quoted everywhere. The study behind it, by 55 and Deloitte for Google, covered 37 European and American brand sites, more than 30 million sessions and 30 days at the end of 2019. Its findings are funnel steps, not conversions: in retail, product page to basket rose 9.1%; in lead generation, the number of users reaching the form submission page rose 21.6%. The condition matters: every page in the journey had to improve on four metrics (first meaningful paint, input latency, load time and server latency) by 0.1 seconds. It was observational, with no redesigns during the study. One 0.1 second win on one page is not that experiment. The practical reading is to remove the biggest delay before the first screen, not to chase tenths. See the [Core Web Vitals guide](https://lainagency.com/resources/core-web-vitals-website-speed-guide).

## Benchmarks are not targets
Unbounce's Q4 2024 benchmark reports a median landing page conversion rate of about 6.6%, from roughly 41,000 pages, 464 million visitors and 57 million conversions. Those totals divide to 12.3%, nearly double the median, which shows how a few high-volume pages pull the pooled figure up. Unbounce also says a conversion can be defined in many ways. Your own baseline, with your own definition and traffic source, is the only number you can compare against.

## Can you test the change? The arithmetic
Computed for a 5% baseline, two variants split evenly, 5% two-sided significance and 80% power. "Visitors per day" is the page's total traffic.

| Lift you want to detect | Visitors per variant | 100 a day | 500 a day | 2,000 a day |
|---|---|---|---|---|
| +10% (5.0 to 5.5%) | 31,200 | 625 days | 125 days | 31 days |
| +20% (5.0 to 6.0%) | 8,200 | 163 days | 33 days | 8 days |
| +50% (5.0 to 7.5%) | 1,500 | 29 days | 6 days | under 2 days |

With 100 visitors a day for three weeks, the smallest lift you could reliably detect is about **60%** (5.0 to 8.0%). Different calculators vary by tens of percent depending on the method, and Evan Miller's rule of thumb n = 16σ²/δ² gives somewhat smaller numbers. Peeking is the second trap. Stopping a test when it first looks significant inflates false positives: Miller shows that a no-effect test, stopped at the first significant reading or after 150 observations, gave a false positive 26.1% of the time in his example (a worst case, checked after every observation), and that after ten looks you need a reported 1% to be an honest 5%. Fix the sample size before you start.

| Your traffic | What to do |
|---|---|
| Under about 200 a day | Do not A/B test. Make one large change, run five user tests, compare before and after with caution |
| 200 to 2,000 a day | Test only large changes: a new offer, a new first screen, a removed section |
| Over 2,000 a day | Test one element at a time, with the sample size fixed in advance |

The traffic bands are suggestions read off the table above, not a published rule.

## Where landing pages go wrong, by cause
| What you see | Cause | How to check | Fix |
|---|---|---|---|
| Clicks, few conversions | Visitors from several intents land on one page | Split results by ad group or query | One page per intent cluster |
| Good on desktop, poor on phone | The first screen shows decoration, not the offer | The cold test at 390 by 844 | Offer and action in the first screen |
| Slow only for first-time visitors | Banner or script is the largest element | LCP element in a trace with the banner | Make the banner small and non-blocking |
| A test showed a winner, the lift vanished | Peeking, or too few visitors | Was the sample fixed in advance | Re-run with the calculated sample |
| Ads disapproved or limited | Pop-ups, back button interference, slow loading | Destination policy list above | Remove the overlay |

The Omni Agent landing page in our [case study](https://lainagency.com/works/omni-agent) runs in a clear order: the first screen with the app in a laptop frame, one dashboard in four ideas, the capabilities, the workflow studio, three plans and a closing call. That is a sequence from what it is to what to do, which is a decision about this product's visitors, not a template.

On page length, we found no study that settles it. NN/g's advice is to test. As judgement, not evidence: the more risk or cost the visitor takes on, the more answers the page needs before the action.

If you want a landing page built to this standard as part of a [website project](https://lainagency.com/about#website-design), [tell us what the page needs to do](https://lainagency.com/contact).

## Frequently asked questions

### What is message match on a landing page?

Message match means the page continues the promise that brought the visitor. With Google responsive search ads, which rotate up to 15 headlines in varying combinations, match the offer and intent shared by every combination, not one headline, and use one page per intent cluster.

### Should a landing page have a navigation menu?

Often a logo-only header is enough. HubSpot removed all external links on five landing pages and saw lifts of 0 to 4% on three and 16% and 28% on two, but reported no duration or significance. The evidence says exits rarely help, not how much removing them is worth.

### How many visitors do I need to A/B test a landing page?

For a 5 percent baseline, detecting a 20 percent relative lift at 5 percent significance and 80 percent power needs about 8,200 visitors per variant. At 100 visitors a day in total that is about 163 days, so low-traffic pages should test big changes or use user tests.

### Is Google Quality Score a ranking factor for my landing page?

Google says Quality Score is a diagnostic tool and not an input in the ad auction. Its parts, expected click-through rate, ad relevance and landing page experience, reflect auction-time measurements, so use the score to spot problems, not as a target.

### Does a faster landing page really convert more?

The Deloitte and 55 study for Google found that improving four speed metrics by 0.1 seconds on every page raised lead generation users reaching the form page by 21.6 percent. It measured funnel steps across 37 sites, not conversions on a single page, so treat it as direction.

### What is a good landing page conversion rate?

Unbounce reports a median of about 6.6 percent for Q4 2024 across roughly 41,000 pages, but says conversions are defined many ways. The pooled rate of its totals is 12.3 percent, so benchmarks vary widely. Compare with your own past baseline instead.

## Sources and fact check

Checked on 28 Oct 2026.

- [Google Ads Help: About Quality Score](https://support.google.com/google-ads/answer/6167118)
- [Google Ads Help: About Ad Rank](https://support.google.com/google-ads/answer/1722122)
- [Google Ads Help: About responsive search ads](https://support.google.com/google-ads/answer/7684791)
- [Google Ads policy: Destination experience](https://support.google.com/adspolicy/answer/16427615)
- [Nielsen Norman Group: Scrolling and Attention](https://www.nngroup.com/articles/scrolling-and-attention/)
- [web.dev: Largest Contentful Paint (LCP)](https://web.dev/articles/lcp)
- [web.dev: Milliseconds make millions](https://web.dev/case-studies/milliseconds-make-millions)
- [W3C: Understanding SC 1.4.11 Non-text Contrast](https://www.w3.org/WAI/WCAG22/Understanding/non-text-contrast.html)
- [HubSpot: Should you remove navigation from landing pages?](https://blog.hubspot.com/marketing/landing-page-navigation-ht)
- [Unbounce: Average landing page conversion rates](https://unbounce.com/average-conversion-rates-landing-pages/)
- [Evan Miller: How Not To Run an A/B Test](https://www.evanmiller.org/how-not-to-run-an-ab-test.html)
- [DebugBear: Cookie consent banner performance](https://debugbear.com/blog/cookie-consent-banner-performance)