Writing Analyzer: How It Works
A writing analyzer reports the measurable properties of a draft — sentence length distribution, passive constructions, adverb density, repetition and readability. None of these measures quality. All of them point at places worth a second look.
What gets measured
| Signal | What it suggests |
|---|---|
| Long sentences (30+ words) | Candidates for splitting |
| Uniform sentence length | Monotonous rhythm |
| Passive voice density | Possibly evasive or indirect phrasing |
| Adverb frequency | Weak verbs propped up by modifiers |
| Repeated openings | Formulaic structure |
| Readability score | Rough accessibility indicator |
Passive voice is not a fault
The advice to eliminate passive voice is over-applied. Passive constructions are correct and often preferable when the actor is unknown, irrelevant, or deliberately de-emphasised:
- 'The samples were incubated at 37 °C' — scientific convention, and the actor genuinely does not matter.
- 'The account was compromised on 14 March' — the actor is unknown.
- 'Payment has been received' — the recipient is the point.
Passive voice becomes a problem when it hides responsibility: 'mistakes were made' is doing something different from 'we made mistakes'. Flag passives, read them, and keep the ones that earn their place. A target of under about 10% is a reasonable guide for most prose; scientific writing legitimately runs far higher.
Adverbs and weak verbs
Most adverbs signal a verb that is not pulling its weight. 'Walked slowly' is weaker than 'ambled'; 'said loudly' is weaker than 'shouted'. The worst offenders are intensifiers that add nothing measurable — 'very', 'really', 'quite', 'extremely', 'actually', 'basically'. Delete them and most sentences improve.
The exception is adverbs that genuinely change meaning: 'he almost finished' and 'he finished' are different claims.
Sentence rhythm
Variation matters more than average length. A passage of uniformly medium sentences reads as flat even when each sentence is well made. Mixing lengths creates pace, and a short sentence after several long ones carries emphasis by contrast alone. Look at the distribution the analyzer reports, not just the mean.
Readability scores, honestly
Flesch-Kincaid and similar formulas combine sentence length and syllable count. They correlate loosely with comprehension and are trivially gamed — chopping sentences arbitrarily improves the score and can make text harder to follow. Use them as a flag when a score is far outside the range for your audience, not as a target to optimise toward.
What no analyzer can see
Whether the argument holds. Whether the structure serves the reader. Whether anything important is missing. Whether the tone suits the audience. Whether it is true. A draft can score perfectly on every metric here and be worthless, and a piece that violates every guideline can be excellent.
Use the analysis as a checklist for a revision pass, then do the part it cannot: read it as the intended reader, and ask whether it does what it was written to do.