Typography and reading on screen: what research measures

Nobody reads a page the way its designer imagines. The NN/g archive on reading spent decades measuring what people actually do with words on a screen, and what it found is narrower than the advice built on top of it. The visual design topic index gathers the pieces behind this post.

What the research can measure in reading

Typographic research measures two things well: how fast a person reads a known passage, and where the eye lands on a page. Speed is countable and placement is recordable. Understanding, which is what matters most, is much harder to measure.

Attention sets the ceiling. The 2008 review of reading behavior found that visitors to an average page have time for at most 28% of the words, and 20% is the more likely case. It rests on an instrumented browser study by Weinreich and colleagues with 25 people, collected in 2005 and reduced to 45,237 page views.

The estimate runs about 25 seconds fixed plus 4.4 seconds for each 100 words, so half the information only fits on a page of 111 words or fewer. The authors name their limits: no eye tracking, so how the time was spent stays unknown, and the data comes from 2005 (NN/g).

Measurement has a second ceiling, and the 2022 study makes it plain. It tested linear reading with questions afterward, not the scanning people do in practice. The author himself calls the outcome disappointing, because it doesn’t lead to a simple rule.

Legibility against comprehension

A 2015 explainer draws the line that content and design teams tend to blur. Legibility is seeing a character and telling it apart from its neighbors. Readability is the difficulty of words and sentences, usually scored in school years. Comprehension is understanding the message and reaching the right conclusion.

The last one is the goal. The first two are ways to get there.

Legibility can be tested. Measure reading speed against a reference design and treat a clear slowdown as a problem. For readability, aim at the eighth grade for a wide audience and the twelfth grade for a technical one. For comprehension, work in the reader’s language, lead with the inverted pyramid, lower the cognitive load, and shorten the text on a phone.

Font, size, and line height

Font choice does matter, but in a curious way. In the 2022 study the fastest font beat the slowest by 35% for the same person, at 314 words per minute against 232 (the best font for online reading). Garamond held the best average at 312 words per minute. Franklin Gothic took the most first places, 59% of the time, on a worse average of 271. That split suggests a font can help its slowest readers the most.

Personal taste is a poor guide. Readers went 14% faster in the fastest font than in the font they liked best, so asking people what they prefer misses the point. Speed also falls with age, about 1.5 words per minute for every year, close to 11% more time every 20 years. Comprehension barely moved across the sixteen fonts and separated none of them (NN/g).

Size and spacing beat the family name. The 2017 MIT AgeLab study tested single words and found that larger beat smaller, regular width beat condensed, and uppercase beat lowercase, each with statistical significance. Lowercase cost 26% more time than uppercase, and condensed cost 11.2% more than regular. The lowercase penalty grew worse as the size shrank.

The study itself carries the caveat. It used isolated words, and the author records that in a long passage uppercase slows reading down.

The 2024 glossary supplies the vocabulary for all of this: aperture, x-height, line height, kerning, orphan, point, family, and weight. A wide aperture tends to read more easily. Left-aligned text is the safer default. An orphan leaves a fragment stranded and makes the passage uncomfortable to read.

On serif against sans serif the glossary doesn’t hedge. Before digital type the serif had the reputation. On screen today the sans serif matches it and sometimes does better, in part because a serif doesn’t always render cleanly at low resolution.

The F pattern and what it decides

The F pattern describes a scan, not a read. The eye crosses the top, drops for a shorter second pass, then runs down the left edge. The opening lines collect the most attention, and the first words of each line collect the most fixations. It holds in the content area rather than the navigation bar, and it appears on phones and mirrors itself in a right-to-left script.

Writers can put this to work. Put the key point in the first two paragraphs. Write a heading and a subheading that say something. Open a heading with the word that carries the information. Group related items, bold the informative word, name the destination inside the link, keep lists short, and cut whatever doesn’t carry weight.

The scan is even tighter on data-dense product pages, where the numbers do the talking. The same principle applies to product microcopy, where a few words have to land the instruction.

Size under the reader’s control

A 2018 note on variable fonts is a case study, not a user test. It reports its own site analytics: 77% of sessions use a window wider than 1100 pixels, 16% run below 500, and 7% sit between the two. Browser support for variable fonts climbed from 2% to 70% in two years, and 80% of the site’s sessions could use them.

The article offers a way to weigh a change. When an improvement serves roughly 80% of the audience, it can be worth building. If the same change harms the other 20% badly, hold it back. The screen isn’t fully in the designer’s hands, so the decision has to respect the edges (NN/g).

Visual hierarchy before size

Visual hierarchy is the order in which the eye takes in the elements, and it starts before anyone reaches for a larger font. Three tools do most of the work: color and contrast, scale, and grouping.

Hierarchy comes from contrast in value and saturation against the surrounding context, not from color by itself. Grouping comes from proximity with white space and from a common region with a border. The squint test, or a stepped blur, exposes the hierarchy that is really there and reveals one that was never intended. The same balance of size, contrast, and space drives material typography.

Where measurement doesn’t decide

Several of the numbers above carry an expiration date or a scope they can’t escape. Four of them point in different directions, and that tension is worth keeping in view.

Uppercase: the glanceable study measured uppercase 26% faster than lowercase for a single word. The glossary and the study’s own author warn that uppercase slows a long passage. The unit of text is the difference, and context settles it.
The age effect: the 2022 reading study found about 11% more time every 20 years. An earlier NN/g study reported 16%. The author raises two explanations, a real change over time or a different metric, and leaves the question open.
Font choice against how little people read: one article treats the font as a high-stakes decision, while another shows readers take in 20% to 28% of the words. They measure different postures, linear reading against scanning, so they don’t cancel each other.
The age of the data: the 28% and 20% figures come from data collected in 2005 and get reused as if current. Name the year whenever you quote them (how little do users read).

Font and typography decisions keep circling one question: how much of this text will anyone actually consume. The measurements above bound the answer without settling it.