Free Readability Scorer

A 75-year history of readability formulas

Readability scoring began with Rudolf Flesch's doctoral work at Columbia in 1943, formalised as the Flesch Reading Ease in 1948: a 0-100 score where higher means easier. The U.S. Navy commissioned a recalibration in 1975 (Kincaid et al., Naval Technical Training Command Report 8-75) that mapped the same surface-level features (syllables per word, words per sentence) to U.S. school grade levels, this is the Flesch-Kincaid Grade Level baked into Microsoft Word since the early 1990s. Other formulas filled gaps: Robert Gunning's Fog Index (1952) for business writing; SMOG by McLaughlin (1969), adopted by the U.S. Department of Health and Human Services as the gold standard for health literacy; Coleman-Liau (1975) and ARI (Smith & Senter, 1967) use character counts instead of syllables, sidestepping the need to count syllables programmatically. The Dale-Chall formula (Edgar Dale, 1948; revised 1995) uses a vocabulary list of «familiar» words. The newer Lexile Framework (MetaMetrics, 1989) and ATOS (Renaissance Learning, 1999) are corpus-based and used by U.S. schools. All these formulas measure proxies, not understanding; treat results as «readability» not «comprehension».

Target grade levels for different audiences

• Health and medical communication. The CDC and NIH recommend writing at or below a 6th-grade level (Flesch Reading Ease ≥ 70, SMOG ≤ 8). The CDC's Clear Communication Index and the AMA's «Health Literacy» guidelines codify this. Hospital discharge instructions, vaccine information sheets, and patient consent forms regularly miss this target by 4-6 grades.
• News and journalism. The Reuters, AP, and BBC house styles target a 9th-grade level (Flesch Reading Ease 60-70). The New York Times averages around grade 9-10, The Economist grade 13-14, USA Today grade 6-7. Tabloid newspapers and Reddit posts often score below grade 8.
• Legal and government documents. The U.S. Plain Writing Act of 2010 requires federal agencies to write public documents in plain language; PlainLanguage.gov suggests grade 8 maximum. The UK Plain English Campaign (founded 1979) targets grade 9 for consumer contracts. Most insurance policies and EULAs sit at grade 14-18, well above almost any consumer's threshold.
• Marketing and SEO content. Yoast SEO and Surfer recommend Flesch Reading Ease above 60 (around grade 8) for general web content. Buffer analysed their blog and found posts at grades 6-9 had 36% higher engagement than posts at grade 13+. Mailchimp recommends grade 7 for email subject lines.
• Education and textbooks. School textbooks target one grade level below the audience: a 9th-grade biology textbook aims for grade 8 readability so struggling students aren't gated out by language. Common Core Lexile bands (2010) suggest specific score ranges per grade.
• Technical documentation. Tools like Microsoft, Google, and Apple's developer documentation aim for grade 8-10 in tutorial content, allowing higher grades for reference material. The «Plain language» refactor of MDN Web Docs (2018-2020) dropped average grade from 14 to 9.
• Academic writing. Journal articles routinely score grade 14-20+, which is appropriate for the audience but makes them inaccessible to non-specialists. Science journalism (Quanta, Aeon, The Conversation) aims to translate down to grade 10-12.

Where readability scoring genuinely helps

• Patient-facing health content. NHS Digital, Mayo Clinic, WebMD, and Healthline all run readability checks before publishing. Missing the grade-6 target excludes the roughly half of U.S. adults at basic literacy (PIAAC 2017). Hospital readmission rates correlate with patient comprehension of discharge instructions.
• Drafting and revision. The score is a feedback signal during writing, not a publishable metric. Write the draft, score it, find the highest-grade-level paragraphs (usually long sentences or jargon), simplify those, re-score. Hemingway Editor (2014) and Grammarly added grade-level feedback specifically for this loop.
• Translation and localisation. Translation memory tools (MemoQ, SDL Trados, Phrase) score source text before translation to flag complex passages for senior linguists. International organisations like UNESCO and the UN translate at target grades 6-8 to maximise audience reach across languages.
• Accessibility and WCAG. WCAG 2.1 Success Criterion 3.1.5 (Reading Level) is AAA: «supplemental content or a version that does not require reading ability more advanced than the lower secondary education level». Tools like axe DevTools don't yet automate this but content authors check manually using readability tools.
• Government and civic communication. The U.S. Plain Writing Act (October 2010), the EU's Clearer Communication initiative, the UK Government Digital Service (GDS) style guide all mandate plain language. Tax forms, voter information, benefits applications regularly score above grade 14, the score is the litmus test for compliance.
• Curriculum alignment. When selecting reading materials for a specific grade, teachers cross-reference Lexile or Flesch-Kincaid scores against the Common Core ranges. Library catalogue systems (Follett Destiny, Lexile.com) include scores so students self-select books at their reading level.
• SEO and content marketing. Google's helpful-content updates increasingly favour readable content. Yoast, Surfer, Clearscope, and SemRush all include readability scoring. Buffer's content team found Flesch Reading Ease 60-80 correlates with longer time-on-page and lower bounce rates.

Mistakes that make readability scores misleading

• Scoring fewer than 100 words. All formulas are statistical and need a reasonable sample. Scoring a single sentence or a Twitter post yields wild swings. 200-300 words minimum for reliable Flesch-Kincaid; 30 sentences minimum for SMOG (its original spec).
• Trusting a single formula. Each formula has blind spots. Flesch-Kincaid penalises long words harshly; Coleman-Liau ignores word frequency entirely; SMOG rounds aggressively. Reporting three scores and taking the median or range gives a better signal than any one number.
• Ignoring domain-specific jargon. A medical article about «myocardial infarction» scores grade 15 even if the surrounding sentences are simple. Formulas only see word length, not familiarity to the audience. Pair readability scores with a glossary or first-use explanations.
• Over-optimising for grade level. Splitting every sentence into 8 words and replacing every multisyllabic word produces choppy, juvenile prose that's actually harder to read for adults. Target the audience's grade level, not the lowest possible number.
• Applying English formulas to other languages. Flesch-Kincaid is calibrated for English syllable patterns. Spanish, German, Finnish, Japanese all need their own calibrations (Fernández Huerta for Spanish, Amstad for German, Anderson's RIX for general use). Running English formulas on translated text gives meaningless scores.
• Treating the score as comprehension. Readability formulas measure surface features. They cannot detect logical confusion, missing context, technical accuracy, or whether the structure makes sense. A grade-6 article can still be incomprehensible if it lacks coherence.
• Pasting text with HTML or markup. Tags, URLs, code blocks, and special characters skew sentence detection and word counts. Strip markup first (this tool tries to but isn't perfect for complex HTML/Markdown).

More frequently asked questions

Which formula should I trust if they disagree?

Pick the formula calibrated for your domain. For health and patient education, SMOG is the U.S. Department of Health and Human Services' recommendation (it's conservative, tends to round up). For general web content and journalism, Flesch-Kincaid Grade Level matches what Word, Google Docs, and Yoast use, so consistency with editing tools matters. For automated scoring (e.g. a CI lint), Coleman-Liau or ARI are more reliable because they don't need syllable counting (which is approximate in software). When formulas disagree by more than 2 grades, look at the text: outlier scores usually flag specific paragraphs.

Does this work for non-English text?

English-calibrated formulas give meaningless results in other languages because syllable-per-word and word-per-sentence ratios differ. For Spanish, use the Fernández Huerta formula. For German, Amstad or Wiener Sachtextformel. For French, the Kandel-Moles adaptation. For Japanese, Chinese, Korean, the very concept of «syllable» doesn't map; you need character-density and JLPT-level analysis instead. Specialised tools like readability.js have separate language packs.

Why is the Flesch Reading Ease score on a 0-100 scale instead of grade levels?

Flesch's 1948 paper used a 0-100 scale where 90-100 = «very easy» (4th grade), 60-70 = «standard» (8-9th grade), 0-30 = «very difficult» (college graduate). The 1975 Kincaid recalibration translated the same surface features into U.S. grade levels for the Navy, which needed to match readers to manuals. Both formulas use the same inputs (syllables/word, words/sentence) but different output scales. Most modern tools (including this one) report both because comparisons are easier when you can pick your preferred unit.

Can AI writing assistants replace readability tools?

LLMs (ChatGPT, Claude, Gemini) can suggest simpler wording but they don't reliably measure readability, they hallucinate scores, give different numbers each run, and average across paragraphs in ways that hide outliers. Deterministic formulas (the ones in this tool) give the same answer every time and let you correlate edits with score changes. The right workflow: use the LLM to rewrite, then use the formula to verify the target grade level was actually achieved. Hemingway Editor (2014) was an early example of combining suggestions with deterministic scoring.

Is my text sent to any server when I score it?

No. All six formulas (Flesch-Kincaid, Flesch Reading Ease, Gunning Fog, Coleman-Liau, SMOG, ARI) run in your browser. Open the Network tab in DevTools while typing or pasting; you'll see zero outbound requests. Safe for medical drafts, internal corporate communications, unpublished journalism, legal drafts, and anything subject to NDA.