Jakob Nielsen published a finding in 1997 that this series has been operating on as a working assumption without ever citing it directly: “79 percent of our test users always scanned any new page they came across; only 16 percent read word-by-word.” People don’t read web pages. They scan them — breadth-first, picking out words and sentences, not top-to-bottom, front-to-back.

The natural question, twenty-nine years later, is whether that still holds. NN/g answered it themselves, and it’s worth reading the answer directly rather than assuming.

Still True, Checked Twice

NN/g re-ran the underlying eyetracking research repeatedly — 2006, 2009, 2013, 2016, 2017, and a fresh large-scale study in 2019 — and published a direct comparison in 2020: “we’ve been saying this since 1997: people rarely read online — they’re far more likely to scan than read word for word. That’s one fundamental truth of online information-seeking behavior that hasn’t changed in 23 years.” The 2019 study went further than the original, deliberately testing in Raleigh and Beijing to see whether the finding held across a completely different writing system. It did: “almost every pattern and behavior we discovered in American users was also demonstrated by Chinese users.” Breadth-first scanning isn’t a quirk of English-language web design. It’s closer to how human attention actually allocates itself against unfamiliar text, regardless of the script it’s written in.

That’s a genuinely strong answer to “is this still accurate” — not a single study assumed to still apply, but the same lab checking its own work across thirteen years and two very different writing systems, and getting the same result both times.

Marketese Is a Cognitive Tax, Measured

The part of the original research worth pulling into this series directly is the measured cost of ignoring it. Nielsen tested five versions of the same paragraph — promotional (“marketese”), concise, scannable, objective, and all three combined — and measured actual task performance, not just preference. Concise text alone: 58% more usable. Scannable layout alone: 47%. All three combined: 124%.

The surprising result was the objective-language version, 27% better on its own, because — in Nielsen’s own words — “promotional language imposes a cognitive burden on users who have to spend resources on filtering out the hyperbole to get at the facts.” His example: a paragraph opening “Nebraska is filled with internationally recognized attractions” produces an immediate, measurable no, it’s not in the reader’s head, and that flicker of dismissal is the actual cost — attention spent discounting the claim rather than absorbing the content underneath it.

That’s the same tax this series has already named from a different angle — the inmates who can’t see their own jargon, the Def-Push copy that makes the reader do the work of translating enthusiasm into fact. Nielsen’s numbers are the missing citation: it isn’t a stylistic preference that marketese sucks. It’s a measured 27–124% usability cost, replicated, over two decades.

New Patterns, Same Underlying Behavior

The genuinely interesting update in the 2020 research isn’t that scanning stopped being true — it’s that new shapes of scanning appeared as the Gutenberg layer (page layout complexity) evolved underneath the same Semantic behavior (people still don’t want to waste effort).

Zigzag and comparison-table layouts produced what NN/g calls the lawn-mower pattern — eyes sweeping left to right across a row, dropping down, sweeping right to left across the next, back and forth like an actual lawn mower, adapting the old scanning instinct to a new visual structure that didn’t exist in the same form in 1997.

Rich Google search results produced something sharper: the pinball pattern — a highly nonlinear bounce between SERP features, images, and result snippets. NN/g’s own example: one participant fixated on 158 separate spots in fourteen seconds searching for Botox pricing. That’s not disorganised attention. It’s the scanning instinct adapting to a results page that stopped being a simple list and started being a wide Gutenberg pipe the user has to narrow down themselves, feature by feature, because the search engine stopped resolving a single ranked answer and started presenting several candidate answers at once.

Neither pattern replaces scanning. Both are scanning, reshaped by a more complex layout, which is exactly what you’d predict if the underlying behavior is genuinely a constant and the page structure is the variable. The F-pattern, the layer-cake pattern, and five other gaze patterns from the original research all reappeared unchanged in the 2019 study, sitting alongside the two new ones.

Breadth-First Was Never Optional

This series has used “breadth-first” as a working discipline — check new material against what’s already published before drafting, not after — treating it as a sensible editorial habit. Nielsen’s research reframes it as something closer to a fact about the reader on the other end: they were always going to read this way regardless of how the post got written. A post structured for linear, front-to-back reading is optimising for the 16% of readers who actually read that way, while the other four in five are scanning for headers, bolded terms, and the first few words of every paragraph to decide whether the rest is worth their time.

That’s the practical upshot worth keeping: subheadings that carry real information rather than cleverness, one idea per paragraph because a second idea buried after the first sentence won’t be seen, and the conclusion stated early rather than built up to. Not because scanning is a lesser way to read. Because it’s the actual, measured, twice-confirmed way almost everyone reads, and writing as though it isn’t is optimising for a reader who mostly doesn’t exist.

This post is part of the Gutenberg/Semantic series. Related: Nothing Is Confusing to Me on jargon as a cognitive tax the reader pays, and Use-Specialisation on staying wide until the point where the reader actually narrows in.

Sources: How Users Read on the Web, Jakob Nielsen, Nielsen Norman Group, September 30, 1997; How People Read Online: New and Old Findings, Kate Moran, Nielsen Norman Group, April 5, 2020.