How many users do you actually need to test a landing page?

In 1993, Jakob Nielsen and Thomas Landauer published a mathematical model of usability problem discovery in the proceedings of the CHI’93 conference (Nielsen & Landauer, 1993). Analysing eleven existing studies, they showed that problem detection follows a Poisson process: each additional tester mostly re-finds problems already spotted by earlier ones, with returns diminishing fast. Their calculation produced the number that became a cult figure in interface design — a single tester finds on average 31% of a page's usability problems, five find about 85%, and beyond that, each extra tester costs the same recruiting effort for an ever-smaller gain. A few years later, Nielsen popularised this result as "you only need to test with 5 users," a line that turned into a CRO dogma — including for landing pages. The problem: that dogma answers one specific question, not every question you have before launching a page.

Where the number "5" actually comes from

Nielsen and Landauer's model rests on a simple formula: the number of problems found after n testers equals N × (1 − (1 − p)ⁿ), where N is the total number of real problems and p is the average detection rate of a single tester. With a p of 31% — the average observed across their eleven studies — the cumulative detection curve looks like this:

• 1 tester: about 31% of problems found
• 2 testers: about 52%
• 3 testers: about 67%
• 5 testers: about 85%
• 10 testers: about 97%

An average rate, not a universal constant

That 31% figure isn't a law of physics: it varies with the complexity of the interface being tested, the observer's experience, and how homogeneous the recruited testers are. Nielsen himself noted, in the original paper, that this number was an average drawn from very different studies — not a guarantee that applies as-is to any given product. That's exactly the point a later study would put to the test.

The study that qualifies the rule: Faulkner (2003)

In 2003, researcher Laura Faulkner published an experiment in Behavior Research Methods, Instruments, & Computers that directly tested how reliable the 5-user rule really is (Faulkner, 2003). She had a web application tested by 60 participants, then randomly drew groups of 5, 10, and 20 testers from that full sample to compare what each group size would have revealed. The result breaks the neat regularity of the theoretical model: some groups of 5 testers uncovered 99% of the real problems — others, drawn from the very same pool of 60 people, uncovered only 55%. With 10 testers, the worst randomly-drawn group still reached 80%; with 20, 95%. In other words, 5 testers are enough on average, but the variance around that average is wide: a poorly drawn group of 5 can miss nearly half of a page's real problems.

Two different questions, two different rules

The most common mistake is applying the 5-user rule to a question it doesn't cover. A qualitative usability test — 5 to 10 people navigating the page while you observe their hesitations — answers "where do visitors get stuck, what don't they understand, what makes them hesitate?". That's a question of kind, not frequency: a single tester who trips over an unclear message reveals a real problem, whether or not the other four understood it. An A/B test answers a completely different question — "which version converts better, on average, across all traffic?" — and that one is a question of frequency: it requires enough visitor volume to reach statistical significance, typically several hundred conversions per variant. The two methods aren't interchangeable: a 5-person usability test will never tell you which of two versions converts better, and an A/B test will never tell you why a page fails to persuade. On traffic that's still low, where an A/B test isn't reliable, qualitative usability testing becomes the main tool available — which is exactly the situation for most landing page launches.

How many testers for a landing page, in practice

1. Before development, on a wireframe or clickable prototype: 5 people matching the target persona are enough to catch major message and structure confusions — the cheapest moment to fix them.
2. On the final page, before launch: keep 5 testers per round, but plan two waves of 5 rather than one batch of 10 — fixing issues between the two waves catches problems a single batch would hide, since the fixes themselves can surface new ones.
3. If the stakes are high — a regulated industry, a redesign of a high-traffic page, a large average order value — scale up to 8-10 testers: that's the threshold where, per Faulkner, even the worst-case draw stays above 80% detection.
4. Once the page is live, complement qualitative judgment with quantitative measurement: a heatmap confirms the friction points found in testing against real traffic, and an A/B test settles the question between two options when qualitative intuition isn't enough.

The 5-second test, an even lighter format

For an ultra-fast check of the message before running a full test, the 5-second test — show the page for 5 seconds, then ask what the visitor remembers — can be run with just as many, or even fewer, participants: it isolates a single variable (how immediately clear the message is) rather than the whole journey.

Running a 5-person test, concretely

• Recruit from the real target audience, not your own circle: friends or colleagues already familiar with the product are usability testing's number-one bias, well before sample size matters.
• Give a task, not a question for opinion: "find the price," "explain in your own words what this page offers" surface blockers that "what do you think?" politely hides.
• Have people think aloud while navigating, to capture hesitation the moment it happens rather than reconstructed afterward.
• Note the silent friction points too — a visitor who hesitates three seconds before clicking, without saying a word, reveals just as much as an explicit comment, and it's often tied to poorly managed cognitive load on the page.
• Change only one variable between two test waves, otherwise there's no way to know which fix solved which problem.

Starting from an already-proven structure shrinks the surface you actually need to test: our 10 LanderKit templates (€89 each, €229 for the full pack) apply page patterns validated across many client contexts, from a SaaS waitlist launch to a coach's booking page — what's left to test is what's genuinely specific to your offer: the copy, the hero image, the argument, rather than the page's entire structure.