Cutting the words in half made the page 58% more usable.

The finding that gets quoted is "users scan, they do not read". The finding that is actually useful is that the study separated three changes and measured them one at a time.

Concise meant roughly half the words. Scannable meant bullets, headings, bold keywords and shorter paragraphs. Objective meant taking out the promotional language — the marketese. Each was tested on its own.

A follow-up applied all of it to a real site. Word count came down to 54% of the original and three long pages became eight short ones averaging 346 words against 2,232. Measured usability came out 159% better, task time 80% better and errors 809% better.

The 27% is the uncomfortable one

Concise and scannable are easy to agree with. Nobody defends long paragraphs.

Objective is different, because marketese is what most professional firms believe their website is for. Removing "world-class", "seamless" and "industry-leading" feels like removing the selling. The study says it is worth 27% of usability, and the mechanism is not subtle: a claim with no evidence attached costs the reader attention and returns nothing, so they discount the next sentence too.

The replacement for an adjective is a number. "Significant rework" becomes "35% of cases come back for corrections". One of those can be checked, which is why one of them persuades.

What the 2017 update changed

Nielsen Norman Group's 2017 eye-tracking update did not retire the F-pattern — it appears when text is unformatted, the reader wants efficiency and motivation is low — but named five other patterns beside it: layer-cake, spotted, marking, bypassing and commitment. Two of those are directly actionable.

Layer-cake is a reader moving down headings and skipping the text between. It means headings have to carry the argument on their own, and it is why NN/g says to start a heading with the words carrying the most information.

Bypassing is a reader skipping lines that begin the same way. Three paragraphs opening "We also" costs you two of them. This is the rule almost nobody knows and it is trivially checkable.

The framing matters more than the list. NN/g argues scanning is rational: readers optimise their return across the whole web, not on your page. You are not being skimmed because the writing is poor. You are being skimmed because that is the correct strategy, and the only response is to be worth stopping for quickly.

We ran this on ourselves and failed

This week we turned eight of these rules into a build check and ran it across 37 articles that had already shipped. It found 18 faults: nine paragraphs over 90 words, seven bolded sentences, and two links whose text was the word "here".

The first threshold we set for bold failed 18 of 37 articles, which did not mean the articles were wrong — it meant the rule was. This site bolds every figure on purpose, so measuring density caught the house style rather than a fault. What actually goes wrong is a bolded sentence, because past a dozen words the emphasis stops pointing at anything.

All 18 are fixed and the check runs on every build, which is the only version of a style guide that survives a busy fortnight. The whole standard is written down in the repository beside the code that enforces it.

What this does not tell you

The study is from 1997 and the web has changed enormously. What has not obviously changed is the reader's incentive, and the 2017 eye-tracking work found the same scanning behaviour on modern pages — but we would not claim the exact percentages transfer to 2026. Treat them as evidence that the direction is real and large, not as a calculator.

It also says nothing about whether anyone arrives. A perfectly concise page nobody reaches is still nobody reading, which is the argument in the note on being invisible to AI search and the reason a crawler default can matter more than a rewrite.

It says nothing about whether the thing you are writing is true — the register of what we checked and the nine figures we removed are the other half of that. If you would like the same discipline applied to your pages, the drafting work starts there.