Text Tools · runs in your browser

Writing Variance Analyser

Find where your writing has fallen into a pattern — uniform sentences, stock transitions, worn phrasing

No sign-up No file limit Nothing uploaded

Writing Variance Analyser

How to use 3 steps
  1. Paste under Your writing.
  2. Press Analyse Writing.
  3. Review how sentence length and rhythm vary, and Copy Report to keep it.

A few paragraphs works better than a single one. Nothing is uploaded — the text is analysed in your browser.

Writing Variance

How this tool works

The main measurement is how much sentence length varies. Prose that swings between short and long holds attention; prose where every sentence lands within a few words of the last reads flat, however good each sentence is on its own. Alongside that, the report names the sentences that all begin the same way, the stock transitions, and the phrasing that has been worn thin by overuse.

Every finding points at a place in the text rather than producing a score, because a number tells you nothing you can act on. Sentence four opens the same way as sentences five and six; sentences eleven to fifteen are all within a few words of each other; this phrase appears six times. Those are edits you can make. The longest sentences are listed last, because splitting one is usually the single change that improves the rhythm most.

What this is and is not

  • It is not an AI detector, and does not report a probability: those do not reliably work — OpenAI withdrew its own for poor accuracy — and they misfire far more often on people writing English as a second language. A percentage would be inventing certainty that nobody has, and being accused on the strength of one is a real cost to a real person.
  • Uniform writing is worse writing whoever wrote it: that is why this measures the shape of the prose rather than guessing at its origin. Text that genuinely varies also reads as less machine-made, but as a consequence of being better rather than as a trick.
  • A short sentence is what gives a long one its weight: if nothing in a passage is under nine words, that is usually the quickest fix available. One deliberately blunt sentence changes the rhythm of everything around it.
  • The flagged phrases are not errors: delve, leverage, seamless, robust and the rest are ordinary words that have been used so heavily they now read as filler. Use them where they are right; notice when they are doing no work.

What it measures instead of guessing at authorship

The core number is how much sentence length swings across a passage — the standard deviation of words per sentence. Prose that alternates between short and long holds attention; prose where every sentence lands within a word or two of the last reads flat no matter how well any individual sentence is built. Around that central measurement, the [msm_tool slug="writing-variance"] separately flags sentences that open the same way, transitions repeated past the point of usefulness, and a specific list of phrases — delve into, leverage, seamlessly, robust, cutting-edge, a testament to, a game changer, and others in the same worn family — that show up so often they have stopped doing any work.

Why every finding points at a place, not a percentage

Nothing here produces a single score, because a number alone tells you nothing you can act on. The output instead reads like sentence four opens the same way as five and six, or this phrase appears six times, which is an edit rather than a verdict. The longest sentences in the passage are listed last on purpose, because splitting one is usually the single change that improves the rhythm of everything around it.

Deliberately not an AI detector

This tool does not report a probability that something was AI-written, and that is a decision rather than a missing feature. Detectors that attempt it do not work reliably — OpenAI withdrew its own for poor accuracy — and they misfire hardest against people writing English as a second language, whose sentence patterns are often more regular for reasons that have nothing to do with authorship. A percentage invents a certainty nobody actually has, and someone wrongly accused on the strength of one has no way to disprove it.

Uniform writing is worse writing regardless of who or what produced it, which is the reasoning behind measuring shape instead of guessing at origin. Genuinely varied prose does read as less machine-generated, but that happens as a side effect of being better writing, not because variation was added as a trick to defeat a detector — and padding in synonyms or invisible characters to fool one would not pass this analysis anyway, since it measures rhythm and repetition rather than word choice alone.

What counts as varied, and what counts as flat

As a rough guide, deviation above roughly nine words reads as strongly varied, and under about three reads as flat. Both are guides rather than targets: technical writing is legitimately more uniform than an essay, and rewriting a passage to hit a specific deviation figure is the same mistake as padding an article to hit a word count.

Why it asks for more than a sentence or two

Variation needs several sentences to vary across. Three is the minimum accepted, and a few paragraphs produce a figure worth trusting — short of that, one unusually long or short sentence distorts the whole measurement on its own.

Nothing leaves the browser

The analysis runs entirely client-side, so pasting an unpublished draft here for a check does not send it, or any part of it, anywhere.

Frequently Asked Questions

Specific to the Writing Variance Analyser. Broader questions about this category are answered on the category page.

No, and no tool reliably can. Detectors produce false positives at a high rate, particularly for people writing in a second language, and someone accused on the strength of a score has no way to disprove it. This reports the shape of the writing and leaves the conclusion to you.
Indirectly, and only by actually improving the writing. Detectors mostly react to low variation, so text with real rhythm scores lower. What this will not do is swap in synonyms or hide invisible characters, which makes writing worse and is dishonest besides.
Above about nine words of deviation reads as strongly varied, and under three reads as flat. Those are guides rather than targets — technical writing is legitimately more uniform than an essay, and writing to hit a number is the same mistake as writing to hit a word count.
Variation needs something to vary across. Three sentences is the minimum it will accept and a few paragraphs gives a figure worth trusting; below that a single long sentence distorts everything.
No. The analysis runs entirely in your browser, so pasting an unpublished draft here does not send it anywhere.

Subscribe to receive notifications about new tools, features, and productivity tips.