Readability Metrics That Matter (and Ones That Don't)

If you paste your homepage into a readability calculator, it will give you a number. Flesch-Kincaid grade level, maybe, or a Gunning Fog index. The number will feel scientific. It is not entirely wrong, but it is measuring something narrower than you think, and it can be gamed in ways that make your writing worse.

Most of these formulas are built on two ingredients: how long your sentences are and how long your words are. That’s it. They were invented decades ago to grade textbooks and military manuals, and for that job they work reasonably well, because in ordinary prose long sentences and long words really do correlate with difficulty. But correlation is doing all the work in that sentence. The formulas don’t know what your words mean. “Utilize” and “banana” score the same. A sentence of pure jargon made of short words scores as easy. A clear sentence that happens to run long scores as hard.

The failure mode is that people optimize the number instead of the reader. They chop every sentence in half. They swap precise words for vague short ones. The score improves and the writing gets worse, because rhythm and precision were carrying meaning that the formula couldn’t see. This is Goodhart’s law applied to prose: when a measure becomes a target, it stops being a good measure.

So what actually matters? I built GazeSite partly to answer this question honestly, and the honest answer is that the metrics that predict whether someone reads your page are mostly not about individual sentences at all. They are about the shape of the page.

The first one is what I’d call time-to-point. How many words does a visitor have to read before they know what the page is about? On most sites the answer is depressingly large. There’s a clever headline, then a paragraph of throat-clearing, and the actual point arrives in paragraph three. On a phone, paragraph three is two screens down. Nobody who doesn’t already care will get there. This is measurable. Count the words before the first sentence a stranger would find informative. If it’s more than about thirty, you have a readability problem no grade-level score will detect.

The second is paragraph length as it renders, not as it looks in your editor. A paragraph that looks modest on a 27-inch monitor becomes a wall on a 390-pixel phone screen. Text reflows; a hundred words that filled four lines on desktop fill ten on mobile. The metric that matters is lines-per-paragraph at mobile width, and almost nobody measures it because almost nobody writes at mobile width. When I audit pages, this is the single most common problem: prose that was readable where it was written and unreadable where it is read.

The third is typographic comfort, which sounds fuzzy but reduces to a few numbers. Font size at or above sixteen pixels on mobile. Line height around one and a half times the font size. Line length somewhere between forty-five and seventy-five characters. Contrast that passes the WCAG threshold, which for body text means a ratio of at least 4.5 to 1 between text and background. These are boring, checkable facts, and violating any of them makes reading physically harder in a way no amount of good writing compensates for. Light gray text on white is the classic case. The prose might be brilliant. Nobody will squint long enough to find out.

The fourth is scannability, which you can measure by an odd test: delete everything except the headings, link text, and bolded phrases, and read what’s left. Does it tell the story? Real readers skim first and read second, if at all. The skeleton of your page is the version most people actually consume. If the skeleton is generic — “Features,” “About,” “Learn more” — the skim tells them nothing, and they leave before the full text gets its chance.

Notice what’s absent from this list: syllable counts, adverb counts, passive-voice percentages. These aren’t useless as editing prompts. If a tool flags a passive sentence, looking at it costs you nothing. But they’re diagnostics, not goals. Treating them as goals produces prose that scores well and reads like it was written by someone following rules, which is exactly what it is.

There’s a deeper point here about measurement in general. The metrics that matter are usually the ones closest to the actual behavior you care about, and the metrics that don’t are the ones that were merely easy to compute. Word length was easy to compute in 1948, so that’s what the formulas used. Rendering a page on a phone and asking whether a distracted stranger would keep reading was not easy to compute, so nobody did. Now it is, which is roughly what GazeSite does when it grades readability: it looks at the page the way a reader encounters it, not the way a formula parses it.

If you want one practical rule, it’s this. Read your page on your own phone, standing up, as if you’d never seen it. Notice the exact moment your attention flickers. That moment is your worst metric, and no calculator will find it for you.