Home › Tools › Readability Score Checker

Readability Score Checker

Flesch Reading Ease, grade level and the sentences dragging you down — calibrated on 46 real homepages (median 57), plus when to ignore the score.

By Hongtao Ren · Updated 2026-09-22

Paste your body copy for a readability breakdown — the standard scores plus the specific sentences that are hardest to read. Calibrated against 46 real homepages so you can tell whether your number is actually unusual.

How to use it

1. Paste the body copy only — not the navigation and footer.
2. Read the Flesch score and grade level together — one without the other is easy to misread.
3. Open the longest sentences list and split anything over about 25 words.
4. Check words per sentence. Under 8 means you scored fragments, not prose, and the number is not telling you what you think it is.

How to read the two scores together

Flesch Reading Ease runs from 0 to 100, higher being easier; 60 to 70 is plain English. Flesch-Kincaid grade level expresses the same signal as US school years. A text can score 65 (fine) and grade 9 (also fine) — they agree more often than not.

Both formulas are built from exactly two inputs: sentence length and syllables per word. That means there are only two levers available to you — shorter sentences, and shorter words.

What 46 real homepages actually score

We ran this exact formula — the same syllable counter this page uses — over the body copy of 78 well-known homepages on 5 October 2026. Six came back in Chinese or Japanese because our crawl exits in Asia, seven had under 100 words of server-rendered text, two were bot walls and fifteen did not return 200. That left 46 pages and 36,733 words of body copy.

The median Flesch Reading Ease was 57.1, with the middle half of the sample between 49.6 and 64.0. Just over half — 26 of 46 — sat below 60, which is to say below 'plain English'. Only 7 of 46 reached 70 or above. The easiest page in the sample was cloudflare.com at 76.0; the hardest was wikimedia.org at 22.5, followed by kubernetes.io at 38.6 and coursera.org at 40.0.

Grade level is where the usual advice gets strange. The median was grade 6.7 and the highest in the entire sample was 11.1. Not one page reached grade 12. So 'write for grade 8' is not an aspirational target — it is roughly what large commercial sites already do, and a page at grade 11 is at the edge of what any of them publish.

The categories differ more than the pages inside them. Ecommerce and finance homepages came in around 70 (easy, short words, short lines); developer and cloud infrastructure sites sat near 44–53 (product names and technical nouns); education and government came in around 49–57. If your page scores 50 and it is documentation for developers, you are normal. If it scores 50 and it is a landing page selling something to everyone, you are an outlier and worth rewriting.

The score moves when the formatting moves

This is the single biggest problem with readability scores, and it is invisible unless you go looking for it. Both formulas divide by the number of sentences — and 'how many sentences is this' has no fixed answer.

Take a homepage made of headlines, buttons and one-line feature blocks. Count every block as a sentence and the average sentence is short. Count only full stops and the same page becomes a handful of enormous sentences. We measured both ways on the same 46 pages: median words per sentence was 5.9 counting blocks versus 22.1 counting full stops. Median sentence count was 125 versus 34.

That gap turns into a median 16.8-point difference in the Flesch score. On 34 of 46 pages the two methods differed by 10 points or more, on 15 pages by 20 or more, and on who.int by 124.6 — because its homepage is almost entirely news headlines with no terminal punctuation at all.

This tool splits on line breaks as well as full stops, so a paste with paragraph breaks scores closer to the block count. That is the right choice for prose, and the reason you should never compare your number against someone else's tool without knowing which rule they used. Test it yourself: paste the same text once with real paragraphs and once as a single unwrapped block. If the score moves more than about 10 points, you were measuring your formatting, not your writing.

Paste the body, not the page

Navigation menus, footers and forms are visible text, and they are not prose. Across the 46 pages, boilerplate accounted for a median 34.8% of all visible words — and on trello.com it was 86.3%, with forbes.com at 76.2% and vercel.com at 72.1%.

Because menu items are short words with no sentences, they usually pull the score in both directions depending on the page. Excluding them moved the Flesch score by a median of only 2.8 points, but the range ran from −21.6 to +16.4, and on 6 of 46 pages it flipped which side of the 60 line the page fell on: shopify.com went from 57.3 to 70.4, forbes.com from 52.3 to 63.1, and trello.com the other way, from 64.4 down to 42.8.

So: paste the article body. If you score a whole rendered page and act on the number, there is a real chance you are reacting to the menu.

Where these formulas mislead

Syllable counting is a heuristic, and it treats every polysyllabic word as equally hard. 'Photography' and 'responsibility' are not equally difficult, but both cost the same in the formula. Technical vocabulary is necessary in technical writing and should not be removed to chase a number.

Use the score to spot sentences that are structurally tangled, not as a target to hit. A page at grade 12 about distributed databases is not failing; a page at grade 16 about choosing a kettle probably is.

The three edits that actually move the number

Split sentences at conjunctions. Most sentences over 30 words contain an 'and', a 'but' or a 'which' that marks a natural seam. Cutting there produces two sentences that read better than the original and need no rewriting of the ideas.

Replace nominalisations with verbs. 'The implementation of the configuration requires the specification of parameters' is four nouns doing a verb's work. 'To configure the tool, specify its parameters' says the same thing in a third of the words and drops the syllable count sharply.

Delete the throat-clearing. Phrases like 'it is important to note that' and 'in order to' add syllables without adding meaning. Removing them is the cheapest readability win available, and it almost always improves the writing rather than just the score.

Do those three things and the Flesch score moves on its own. Rewriting specifically to raise the number — swapping 'use' for 'utilise' in reverse, or breaking sentences at arbitrary points — produces text that scores well and reads worse.

When not to use a readability score at all

Do not score text in a language the formula was not built for. Flesch and Fog were derived from English prose; the syllable rules are English syllable rules. In our crawl, six of 78 pages came back as Chinese or Japanese editions — stripe.com, linkedin.com, paypal.com and spotify.com among them — because the requests exited in Asia. We dropped them rather than publish a number. If you paste Chinese or Japanese text here, the output is arithmetic, not a measurement.

Do not score very short text. Ten of the 46 usable pages had under 300 words of body copy, and those are exactly the pages where a single long sentence or a stray product name swings the result. Below about 300 words, treat the score as a rough gesture.

Do not score lists, tables of contents, navigation, product grids or pages built from headlines. The formulas assume sentences; fragment collections have none, and the score you get back is an artefact of how the fragments happen to be split.

And do not treat the score as a goal in itself. Every lever that raises it — shorter words, shorter sentences — is available to a writer who has nothing to say. Use it to find the paragraph that made you work, then trust your own reading of that paragraph over the number.

How we measured the 46-page sample

The domain list is the same one behind our heading structure and favicon surveys, so the numbers are comparable across those three. For each site we fetched the homepage, stripped script, style, template, noscript, svg and iframe, then removed everything inside nav, header, footer, aside and form elements using tag-stack matching rather than a text heuristic. Body copy is what was left.

We counted syllables with the same function this page uses, loaded from the same file, so the survey and the tool cannot drift apart. We excluded bot walls by matching challenge-page text, and dropped pages whose body copy came back in a non-Latin script. The script is in the repository as scripts/survey-readability.mjs and re-runs offline with --report.

Two things it cannot see: content injected by JavaScript after load, which is why client-rendered homepages like reddit.com and airbnb.com came back with almost no text and were excluded rather than counted as unreadable; and the fact that 78 hand-picked well-known domains are not a random sample of the web. The numbers describe famous homepages, nothing else.

Related tools