Jake Young
Your next visitor has probably already read about you. An AI assistant summarised your category, named a few options and answered the basic questions before they clicked. Then they landed on a first screen that explains what the assistant just told them, found nothing they couldn’t have guessed, and left without feeling anything at all.
The sites people remember use those same ten seconds differently. They make a promise specific enough to follow, hold back something worth scrolling for, and deliver one moment that sticks. That can be designed on purpose, with four ideas from decades of research and a teardown you can run on your own site this week.
The Visit Now Starts Halfway Through
Pew tracked the browsing of 900 US adults and found people clicked a traditional search result in 8% of visits when a Google AI summary appeared, against 15% when none did. Our essay Designed to Be Cited covers how to be the source those summaries quote. This one is about the people who still click through. They know roughly what you do. What they came for is conviction: is this the one, and do I want it?
Time is short, and so is scroll. Liu, White and Dumais studied dwell times on more than 200,000 pages and found the first ten seconds largely decide whether people stay, while those who pass thirty seconds often stay for minutes. Nielsen Norman Group’s eye tracking found 57% of viewing time went to the first screen and 74% to the first two. A page has to earn every screen after that.
Scent, Gap, Peak, End
We plan attention with four ideas, and every screen of a page should be doing at least one of them.
Scent: every screen promises the next
Information foraging theory, developed by Peter Pirolli and Stuart Card at Xerox PARC, describes people moving through information the way animals forage, following cues that promise a good return. Nielsen Norman Group calls those cues information scent. Strong scent is specific. “See how the finish is applied” has it, and “Learn more” has almost none. When scent is weak, people stop moving. When a promise isn’t kept on arrival, they stop trusting your scent everywhere.
Gap: open a question worth closing
George Loewenstein’s information gap theory describes curiosity as the feeling of a gap between what we know and what we want to know, and it peaks when people know a little rather than nothing or everything. So give enough to make the question specific, then answer it one screen later. Progressive disclosure applies the same idea to interfaces: show the few most important things first, reveal more on request, and rarely go beyond two levels.
Peak: one moment built to be remembered
In 1933 Hedwig von Restorff showed that an item standing out from similar items is remembered better. A page where every section has equal weight gives memory nothing to hold, so we choose one moment per page to be the peak: a product turning in the light, a number that reframes the problem, a line nobody else in the category would write. Everything around it is quieter so it can stand out.
End: the last moment colours the whole
Daniel Kahneman and colleagues found in 1993 that people judge an experience largely by its most intense moment and by how it ends. Most websites put their least designed screens at the end: a generic form, a confirmation page with no personality, an automated email written by nobody. The end is where desire turns into action or quietly leaks away.
Pacing a Page as a Sequence
We storyboard pages as a sequence of screens, the way a film editor thinks about shots. Each screen is a beat with one job, and dense beats alternate with quiet ones so attention can rest. The rhythm we use most: a first screen with a specific promise, a second that opens a gap, a few that deliver on it with evidence, a peak around the middle while attention is still high, and an ending that closes the loop with one clear action.
Within each screen, hierarchy does the same work. Size, contrast and position decide what the eye takes first, second and third, and a screen with no clear first thing asks the visitor to do the editing.
Two pitfalls come up constantly. The first is the false floor, a full width band or oversized hero that makes the page look finished, so people never scroll. The second is confusing withholding with disclosure. Hiding what people came to check, such as how pricing works, creates suspicion rather than curiosity. The two look identical in a wireframe, so we test: a gap is good if the visitor wants the answer, and bad if they need it and can’t find it.
What Separates Memorable Sites From Competent Ones
Research on first impressions, such as Tuch and colleagues’ 2012 study, found that pages which look typical for their category make better first impressions. That’s useful and also a trap, because a typical page is easy to trust and hard to remember. Generative tools deepen the trap by producing the most typical version of whatever you ask for. Memorable sites use a familiar structure so people navigate without thinking, then spend their distinctiveness on things people can recall later.
Jenni Romaniuk’s work on distinctive brand assets gives the test: an asset is distinctive when people link it to your brand and not to competitors. A colour used with conviction, a typeface nobody else in your category uses, a recurring visual device, a way of writing. The other difference is a point of view. A competent site describes a service, while a memorable one argues for how the work should be done, which gives visitors something to agree with and repeat. Our essay Taste Is the Last Moat explains why that judgment is now hard to copy.
Walkthrough: A Teardown You Can Run on Your Own Site
This is the teardown we run before redesigning anything. It takes about a day for one page, and it works on competitors too.
Step one: capture the sequence. Screenshot every screen of the page at 390 pixels wide and at 1512, and lay them side by side as a film strip. Seeing the page as a sequence exposes pacing problems that scrolling hides.
Step two: run the ten second test. Show five people the first screen for ten seconds, then ask what the company does, who it’s for, and why they’d choose it. If the answers are vague or wrong, nothing further down matters yet.
Step three: audit the scent. List every link, button and heading, and mark any label that could sit on a competitor’s site unchanged. Follow each one and check the destination keeps its promise.
Step four: map the gaps. For each screen, write the question it opens and where that question is answered. A screen that opens nothing is flat, one that opens five questions is noise, and a question never answered is either clickbait or a hidden fact.
Step five: check the disclosure layers. Everything a first time visitor needs to decide belongs in the first layer, with detail one click away and never more than two levels deep.
Step six: find the peak. Ask the same five people which moment they’d describe to a friend. If nobody names one, designing one is the most valuable change you can make. If they all name different ones, the page needs editing.
Step seven: walk the ending. Complete the main action yourself, read the confirmation page, open the automated email, and note how long a real reply takes.
Step eight: test memory. A day or two later, ask your testers what they remember, unprompted, and compare it with what you wanted them to remember. It’s the measure that matters most and the one almost nobody takes.
Then add two checks for the AI era. Paste the page’s text into an assistant and ask for a three sentence summary. If it could describe five competitors, the copy is generic, however good the design looks. And ask a few assistants to describe your company, since that answer is now many visitors’ first impression.
What AI Changes About Attention
Assistants answer the informational questions, so visitors arrive later in the decision with sharper questions. Shopping and research agents compare on attributes and never feel a peak or an ending, which is why, as we argue in Your Next Customer Is an Agent, the website is splitting into a data surface for machines and a conviction surface for people. This essay is about the second.
Generative tools also make a competent page nearly free, which makes a memorable one relatively more valuable. The decisions that create memory, such as which moment becomes the peak, how much to hold back and what the page argues, are taste and strategy. Ask a model and you get the most average answer, and average is exactly what gets ignored.
Our prediction: the sites that keep their traffic will look less like brochures and more like experiences built for the moment of decision, with fewer pages, stronger peaks and far more care over endings.
What to Do This Quarter
Run the teardown on your homepage and your most important conversion page, and do the memory test even if you skip everything else.
Rewrite every “Learn more” as a specific promise. Choose one peak per page and quieten what surrounds it. Redesign the ending, from the form to the time it takes a real person to reply.
Rewrite the first screen for someone who has already read an AI summary of your category. Assume they know what you do, and show them why you.
If your site is being skimmed and forgotten, we’d love to help. We design websites with a sequence, a point of view and a moment people remember. Get in touch and we’ll start with a teardown of the page that matters most to you.
