What the 5 second test on your website actually proves
Two studies checked what people take from a page in five seconds. Liking is settled almost instantly. Understanding is not, and a page lands about one idea however long anyone looks.
The 5 second test on your website shows a stranger your homepage for five seconds, hides it, then asks what you sell, and the answer is usually not the one you wanted.
Most write-ups stop there and tell you to simplify. The better question is whether the test measures your page or the clock. Two studies checked, and they disagree in a way that turns out to be the point.
How do you run a 5 second test on your website?
Take a screenshot of your homepage. Show it to someone outside your company for exactly five seconds, then hide it. Ask what the company sells, who it is for, and what they would do next. Write the answers down word for word, before you explain anything.
- Use a screenshot, not the live site. A live page invites scrolling, which is a different test.
- Use five strangers, not five colleagues. Anyone who has seen a draft is disqualified for good.
- Ask the specific question first. Memory decays while people answer, so "what do you sell" goes before "what else do you remember".
- Record the words, not the gist. If four people say "software" and your headline says "revenue intelligence platform", that gap is the finding.
Ten minutes, five people, no tooling. The rest of the DIY method is in how to audit your own website.
Want a second opinion on the same page? Run a free Revslip audit alongside it.
Why does a page only land one idea?
Attention runs in two stages. The first is automatic and parallel: colour, shape, layout, the thing that pops. The second is deliberate, one object at a time, and it is the only stage that reads a sentence. Five seconds buys most of the first and little of the second.
That split is why the result feels unfair. Your visitor is not lazy. They run the cheap pass first, and the cheap pass cannot read a feature grid.
A page does not fail because people gave it five seconds. It fails because it asked for thirty.
What does the research on five second testing show?
Liking settles almost instantly and does not improve when people look longer. Understanding does improve with more time. But the number of separate messages a page can land stays close to one, however long anyone stares at it, and that is the finding that should change what you do next.
The sharpest evidence comes from Guillaume Gronier at the Luxembourg Institute of Science and Technology, in the Journal of Usability Studies, 2016. He showed the Luxembourg national weather site to 146 people. One group got five seconds, the other unlimited time. The site had three stated purposes.
- 97.1% of the five-second group named the site's main purpose
- 2.9% named its second purpose, the weather warnings
- 0% named its third, the five-day forecast
- 9.0% and 2.6% is what the unlimited-time group managed on those same two
Read that last line twice. Taking the clock off barely helped, and Gronier found no significant difference between the groups on what the site was for, at p = 0.37. What did move were the judgements that need reading: perceived ease of use rose from 5.03 to 5.58, information quality from 4.96 to 5.36. Design ratings did not move at all, at p = 0.65.
A separate team reached a different conclusion about the clock. Kuric and colleagues, in Behaviour and Information Technology, 2023, ran two, five and ten second conditions across six screenshots and found correct identification of a page's purpose kept climbing between five and ten seconds. Worth naming the interest: that team is UXtweak Research, and UXtweak sells a five second test tool.
The two results are not really fighting. Gronier used one weather homepage chosen for low visual load. Kuric used six screenshots spanning a wide complexity range, and the time effect showed up on the busy ones. Simple page, the clock does not matter. Busy page, it does, which is itself the verdict.
The liking half is older and firmer. Lindgaard and colleagues at Carleton University found visual appeal ratings at 50 milliseconds correlated closely with ratings at 500 milliseconds. People decide whether they like your page before reading a word of it, and more time does not change their mind.
What founders say after they fail it
The reaction is rarely about the design. "They said we were a marketing agency." "Nobody mentioned the pricing, it is right there." "She said it looked nice but had no idea what we do." That last one splits exactly what the research splits. The page looked fine. It said nothing.
How do you fix a page that fails it?
Pick the one thing you want a stranger to say back, and build the first screen around only that. Everything competing with it moves below the fold. Then run the test again on the same five-person standard, with five new people.
- Name the outcome, not the category. "Get paid faster" beats "invoicing platform" because one of them is what the buyer wants.
- Cut the hero to one message. If your first screen makes three claims, Gronier's numbers say two are not arriving anyway.
- Put the audience in the subhead. "For Shopify stores doing over 500 orders a month" answers the second question before it is asked.
- Leave one next step visible. Two equally weighted buttons is a choice, and a choice costs a second you do not have.
People also leave for reasons a screenshot never shows, which is what why people leave your landing page covers.
Not sure which message to lead with? Revslip reads your page and ranks what to fix.
How soon should the numbers move?
Give it two weeks of normal traffic, then compare bounce rate and scroll depth against the same weekday span before the change. Clarity fixes show up early because they hit the first seconds of every session, not a late step most people never reach.
To size it first, use the arithmetic we publish for every finding:
monthly visitors × conversion gap × average order value = monthly recovery
A site at 6,000 visitors converting at 1.1%, aiming at the 1.8% its own best page already hits, at an order value of 80 euros, is looking at roughly 3,360 euros a month. That is a ceiling, not a promise. Our sample is 134 audited businesses, and it is biased: these are sites whose owners already suspected something was wrong, which is not a random draw of the web.
When is the 5 second test the wrong tool?
When the page is not meant to be understood in five seconds. A dashboard, a settings screen, a checkout step, a documentation page: all are tools someone reaches mid-task, carrying context you cannot simulate. Gronier is blunt here, noting usability is perceived after the five seconds are up.
Why we cannot run this test for you This test works because the person has never seen the page and runs out of patience. A crawler is the opposite creature. It reads everything, in order, tirelessly. Revslip can count the competing claims on your first screen and time how long the page takes to load them. It cannot be confused by your page, and being confused is the entire measurement here.
Questions that come up after the test
Three things founders ask once the answers are in: the number itself, the sample size, and whether any of it transfers to a phone.
Is five seconds actually the right number?
It was never derived from anything. Christine Perfetti coined the method in the mid-2000s and the five came from practitioner habit. Gronier notes it had no scientific reference at all before his study. Treat five as a convention that suits simple pages, and use ten for dense ones.
How many people do I need?
Five shows you a pattern in what people say, and ten is safer. Either way it is nowhere near enough to prove a conversion change, so do not quote it as one. That is a separate question about testing on low traffic.
Does this apply to mobile?
More so. The same first screen is about a third of the width, so a message that barely survived on a laptop has less room to arrive. Test the mobile screenshot separately rather than assuming it inherits the result.