Where visitors look first on a landing page. And how to make the eyes swoop

Alex · Founder of Magpie AI
Published · 6 min read

In short
Visitors look first at the top left of a screen and at text, and they scan plain unformatted text in an F shape. To get their eyes onto your headline and main button, make those two the largest, highest-contrast things on the first screen, keep them near the top left, and dim competing pictures and faces. A predicted heatmap guesses a typical visitor's first seconds; it doesn't record anyone.
Contents
Top left, and words. That's where eyes went first across the 1,980 screens in the UEyes study (CHI 2023), whatever kind of screen it was, and text caught them more readily than pictures did.
Good news, then, if your headline already sits up there. Less good if your button is a pale outline over on the right next to a big smiling photo, because that's one of the likeliest things on the screen to be missed (and the fixes for it cost almost nothing).
Where the eyes land in the first seconds#
UEyes sat 62 people in front of web pages, desktop apps, phone screens and posters for seven seconds each, and their eyes drifted high and left. The researchers found a top-left bias rather than the pull to the centre you might expect. Text drew more attention than images (they do point out their screens were mostly in English), and on web pages people liked to scan from left to right.
Nielsen Norman Group saw the same lean in ordinary browsing. Its 2017 study followed more than 120 people and over 130,000 fixations on wide monitors, and about 80% of those fixations fell on the left half of the screen, up from 69% in 2010.
Phones are stricter still. UEyes found people looked at fewer elements on a phone and went back to them less often, so whatever wins the first look there pretty much rules the roost. I'd check your button on a phone before the laptop you built the page on, and if the page is slow to show up there at all, speed on mobile comes first.
What the F-pattern shows, and where it stops#
Jakob Nielsen described the F in 2006, from 232 users looking at thousands of pages. One sweep across the top, a shorter one a little lower down, then a run down the left side. It's how people scan a wall of plain text, and a landing page shouldn't hand them one.
Kara Pernice's 2017 update found it still held, on desktop and on mobile. She was specific about when it happens. The text has little or no formatting, the reader wants to be quick, and they aren't committed to reading every word. Nielsen Norman Group calls the result "bad for users and businesses", because people may skip important content simply because it sits on the right.
Give people cues and they scan differently. Pernice lists the other patterns, the layer cake among them, where people read from heading to heading. For a first screen with a clear headline, one short line and one button, I'd take one narrow habit from the F. Start every heading and line with the words that carry the meaning, because the left edge gets read and the end of a line often doesn't.
Faces and pictures pull the eyes#
Faces win. In Cerf and colleagues' 2007 study, people fixated a face with a probability of over 80% within their first two looks, which is about half a second. Those were photos of scenes rather than web pages, so I'm guessing when I say a large face beside a headline would win the same race, though I'd be surprised if it didn't.
Where the face is looking matters as well. Sajjacholapunt and Ball (2014) showed banner adverts to 72 people, and a face turned towards the advert's words raised attention to the whole banner and to its text, while a face staring straight out at the viewer kept the attention on itself. So if there's a person in your hero photo, turn them to the headline or the button and the picture takes the words under its wing.
Most other pictures get ignored. Nielsen's 2010 eye tracking found people skipped big feel-good images that were only there for decoration, and generic stock photos too, but they did study product photos and real people who work at the company. A picture of your product doing its job earns its space. A stock photo of a laptop on a desk just eats into the room for the first screen's words.
How to pull eyes onto the headline and the button#
Start with contrast. Itti and Koch's 2001 review of attention models says "the perceptual saliency of stimuli critically depends on the surrounding context", which in plain terms means a button stands out by being different from what's around it. A black button on a white page pulls the eyes. Put the same black button on a dark photo and it's gone.
Here's a made-up first screen at ledgerly.example, a bookkeeping app, and what I'd change on it.
| On the screen | What it does to the eyes | What I'd change |
|---|---|---|
| A thin, light grey headline, centred | Low contrast, away from the top left | Dark, heavier and aligned left |
| A ghost button outlined in pale blue, "Get started" | Barely differs from the background | Solid fill in the brand's darkest colour |
| A large photo, right, a woman smiling at the camera | Takes the first looks and keeps them | Turn her towards the headline, or show the app |
| A blue promo strip above the header, "50% off today" | Looks like an advert, gets skipped | Move the offer into the line under the button |
The bottom row is what Nielsen Norman Group calls banner blindness. In Pernice's 2018 study anything boxed, animated or brightly treated like an advert got skipped, along with whatever sat next to it, so don't dress your one offer up as an ad.
Then size and place. The headline should be the biggest type on the screen, with the button tucked close under it so the two land in one glance rather than two. After that it's the button's words, since that's what a visitor reads once their eyes get there, and what your button should say is a whole article of its own.
Predicted heatmaps against real eye tracking#
A predicted heatmap is a model's guess at where an average person looks in their first few seconds with nothing in particular in mind. Real eye tracking records where actual people looked (a few dozen to a few hundred of them in the studies above, often given a task). Both are useful. Neither tells you who signs up.
Magpie's free report draws one of these predictions for each first screen, laptop and phone, over the first fifteen looks, about five seconds. It's worked out in code from a model we fitted on the UEyes recordings. On screens it wasn't fitted on, it matches where UEyes's viewers looked at a correlation of 0.50 on web pages and 0.59 on phone screens, where 1 would be perfect, and screen position alone gets 0.46 and 0.54, so place does a lot of the work and the rest is a modest gain. It doesn't detect faces yet.
It flags one thing in red, when the headline and the main button together hold under half of the strongest attention (the red on the heatmap). That threshold is our own judgement of where a first screen's attention ought to go. We haven't measured it against sign-ups, and I'd rather say so plainly.
So treat a heatmap as a little bird's tip about a stranger's first glance, never a recording of your buyers. If the red sits on a photo, a promo strip or a loud row of customer logos, pull it back with the changes above and check again. Then let your real visitors settle it with a test that counts sign-ups.
Questions people ask#
Does button colour matter for attention?#
Only through contrast, really. I'd pick whichever colour differs most from what's directly around the button, then stop fiddling with it and spend the time on the words.
Should the headline be centred or left-aligned?#
Honestly, either can work if it's large and dark. A short centred headline still sits near the top of the screen, which is most of what matters, and plenty of decent pages centre theirs because that's how the design falls. But both of the eye tracking studies above lean left, so if I were starting from a blank page I'd put it on the left and only move it to the middle when the layout really asked for it, rather than because the template happened to come that way.
Sources
- Nielsen Norman Group, F-Shaped Pattern For Reading Web Content (Jakob Nielsen, 2006) (opens in a new tab)
- Nielsen Norman Group, F-Shaped Pattern of Reading on the Web: Misunderstood, But Still Relevant (Kara Pernice, 2017) (opens in a new tab)
- Nielsen Norman Group, Horizontal Attention Leans Left (Therese Fessenden, 2017) (opens in a new tab)
- Jiang et al., UEyes: An Eye-Tracking Dataset across User Interface Types (2024, from their CHI 2023 paper) (opens in a new tab)
- Cerf, Harel, Einhäuser and Koch, Predicting human gaze using low-level saliency combined with face detection (NIPS 2007) (opens in a new tab)
- Sajjacholapunt and Ball, Human faces with averted gaze can enhance advertising effectiveness (Frontiers in Psychology, 2014) (opens in a new tab)
- Itti and Koch, Computational modelling of visual attention (Nature Reviews Neuroscience, 2001) (opens in a new tab)
- Nielsen Norman Group, Photos as Web Content (Jakob Nielsen, 2010) (opens in a new tab)
- Nielsen Norman Group, Banner Blindness Revisited (Kara Pernice, 2018) (opens in a new tab)

About the author
Alex · Founder of Magpie AI
Alex founded Magpie AI and writes about what turns a landing page's visitors into customers, drawing on what its audits find on real sites, from the first screen to the checkout


