Readability scores are one of the oldest tools in computational linguistics — and one of the most misunderstood. They reduce the complexity of prose to a number, which makes them useful for comparing texts at scale, but also easy to misuse.
Here's what the major formulas actually measure, when to use each one, and what the scores mean in practice.
Flesch-Kincaid: The English Standard
The Flesch Reading Ease formula, developed by Rudolf Flesch in 1948 and later revised with J. Peter Kincaid for the U.S. Navy, is the dominant readability metric for English text:
Score = 206.835 − 1.015 × (words/sentences) − 84.6 × (syllables/words)
The formula penalizes two things: long sentences and long words (measured in syllables). The resulting score runs from 0 to 100:
| Score | Level | Example |
|---|---|---|
| 90–100 | Very easy | Comics, basic instructions |
| 70–80 | Easy | Consumer magazines |
| 60–70 | Standard | Plain English target zone |
| 50–60 | Fairly difficult | Academic papers |
| 30–50 | Difficult | Legal documents |
| 0–30 | Very difficult | Scientific journals |
What it measures well: Sentence length and syllable density are strong proxies for cognitive load in English. A score of 65–75 is the sweet spot for most web content — accessible to a broad audience without being condescending.
What it misses: The formula knows nothing about vocabulary difficulty, topic complexity, or coherence. A sentence like "Dog big red run fast jump now" scores excellently — short words, short sentence — but it's incoherent.
LIX: The Scandinavian Alternative
LIX (Läsbarhetsindex) was developed by Carl-Hugo Björnsson in 1968, originally for Swedish. It uses a different approach — instead of syllables, it counts long words (more than 6 characters):
LIX = (words/sentences) + (long words/words × 100)
The raw LIX score is not normalized to 0–100; it runs roughly from 20 (very easy) to 60+ (very difficult). For practical use in a 0–100 readability display, we invert and scale it.
| Raw LIX | Level |
|---|---|
| < 25 | Very easy (children's books) |
| 25–35 | Easy |
| 35–45 | Medium |
| 45–55 | Difficult |
| > 55 | Very difficult (academic/legal) |
Why LIX works for non-English text: Syllable counting is language-specific — English syllable rules don't transfer to Russian, Spanish, Finnish, or most other languages. Character length is language-agnostic. LIX applies consistently across writing systems, which is why it's the better choice for multilingual tools.
TextPurify uses Flesch-Kincaid for English content and LIX for Russian and Spanish.
What AI Text Looks Like on These Scales
AI-generated text tends to cluster in a predictable readability range: 55–70 Flesch, or LIX 35–45. This is neither easy nor difficult — it's the deliberate "neutral professional" register that language models are trained to produce.
The problem is consistency. Human writers vary sentence length naturally, producing rhythm. AI models produce statistically average sentences throughout — each paragraph feels similar in weight and structure, which creates a kind of readability flatness that's distinctive and identifiable.
Optimizing for Readability
Vary your sentence lengths
Short sentences punch. They wake up readers who are scanning. Longer sentences, like this one, allow you to develop a complete thought with qualifications and nuance before landing on the main point. Mix them.
Prefer concrete words over abstract ones
Abstract vocabulary inflates syllable counts without adding clarity. "Facilitate implementation" scores worse than "help build" — and communicates better.
One idea per sentence
Complex sentences that try to carry two or three ideas simultaneously are harder to process, force readers to re-read, and produce low readability scores for good reason: they are genuinely harder.
Target your audience
A 70+ Flesch score isn't always better. If you're writing for lawyers, doctors, or engineers, a score of 45–55 may be exactly right for your audience. The goal is to match the text to the reader — not to minimize score at all costs.
Limitations to Keep in Mind
Readability scores are useful signals, not ground truth. They don't measure:
- Coherence — whether ideas connect logically
- Accuracy — whether the content is correct
- Engagement — whether readers actually want to keep reading
- Domain appropriateness — a high reading ease score in a legal brief may signal imprecision
Use readability scores as one input in a broader quality assessment — which is exactly what TextPurify's Health Score does, combining readability, water density, artifact count, and passage structure into a single composite metric.
TextPurify computes Flesch-Kincaid for English and LIX for other languages, auto-detected from your locale setting.