Does this sound like me? Compare two writing samples, free

By Jack Stovell X · Instagram · LinkedIn · published 2026-10-02

Paste something you wrote and the piece you want to check. The comparison below scores how alike the two voices are, from 0 to 1, and names the habits that match and the ones that moved. Free, no signup needed.

Paste your own writing first: an old email, an essay, whatever's genuinely yours. Then paste the piece you want to check against it (a draft, an edited version, something ghostwritten for you, an AI draft, another writer's sample). The tool renders right below this paragraph and does the rest.

What the comparison measures

Here's the thing: the first piece you paste becomes the baseline. Everything else gets measured against it, not against some abstract idea of "good writing".

The measuring itself happens in two ways. Code counts what can be counted: sentence length and how much it varies, contractions, commas, exclamation marks, semicolons, brackets, question marks, ellipses, the mix of pronouns, article balance, word length. A small model (Claude Haiku, one call per piece) judges the parts that need judgement: formality, humour, expressiveness, confidence versus hedging, opening style, argument structure, rhythm, sentence complexity, clause ordering, specificity. Both feed into one score, weighted, calibrated. Full method's on the voice measurement framework page, if you want the mechanics rather than the result.

You can paste up to three pieces to check against the one baseline. Each needs 120 or more words, or a URL (we'll fetch the page directly). Anything past 3,500 words gets cut to the first 3,500. So don't paste your dissertation. An excerpt does the job.

How to read the result

Three bands, and they're blunt on purpose.

0.80 and above means "reads like the same writer". The measured habits line up closely.

0.55 to 0.79 is drifting: "recognisable, but the habits are slipping". You might see this when an editor's had a heavy hand, or when something started as yours and got smoothed by a tool along the way.

Under 0.55 is off voice: "a different voice on the same name".

Each result names the top differences and the top matches, feature by feature, with both values sitting side by side. Not just a number. You'll see, for instance, that your baseline runs short punchy sentences and the checked piece runs long balanced ones, or that your contraction rate is high and the other piece barely uses any. That's the useful part. The score tells you how far apart two voices are; the feature list tells you why.

Comparing writing styles fairly

Compare like with like. Same kind of writing, similar length, roughly the same job. That is the basis of a fair comparison.

A LinkedIn post against a full essay will drift for reasons that have nothing to do with voice. Posts run short, punchy, front-loaded for a scroll. Essays build. Compare the two and you'll get a low score that tells you about format, not about whether the writing's yours. So match the genre before you trust the number.

120 words minimum, each piece. Shorter than that, and single quirks (one long sentence, one stray exclamation mark) start swinging the score around more than they should.

And if the question is whether a brand stays consistent across everything it publishes, the brand voice consistency checker frames the same measurement for a brand: one main piece, up to three others.

What it cannot tell you

To be fair, this tool has limits, and they're worth saying plainly rather than burying in small print.

It is not proof of authorship. A high score means the habits match; it doesn't mean you can prove who typed what, and it shouldn't be waved around as forensic evidence.

It's not a quality score either. A piece can match your voice closely and still be badly argued, badly structured, or just dull. Voice and quality are different axes entirely.

Short texts are noisy. Under 120 words, one unusual sentence can shift the whole reading. That's exactly why the minimum exists.

And part of the judgement runs through a model, not pure counting. The qualitative half (tone, confidence, rhythm) is read by Claude Haiku, and a model reading prose is not the same as a ruler measuring a plank. It's a considered judgement, not a fixed fact.

From two samples to a full profile

One sample is a snapshot. Useful, but it's a single frame from a much longer film.

A full profile is built from three or more pieces of your writing and measures 45 attributes properly, not just the ones that fit into one comparison. Once it exists, it scores every draft you run against it afterwards, consistently, rather than you pasting two things in every time you're unsure. The first profile is free.

If you'd rather see the raw feature breakdown of a single piece with no model call involved (just the counted, code-side signals) the in-browser style analysis does that on its own, no comparison needed.

Questions

How is this different from a plagiarism checker?

Completely different job. A plagiarism checker looks for copied text; this looks for matching habits: sentence rhythm, punctuation, pronoun use, tone. Two pieces can share zero identical words and still score high here, because voice is about how something's written, not which words got borrowed.

Can I compare two AI drafts against each other?

Yes. Paste one as the baseline and the other as the piece to check. You'll get the same 0 to 1 score and the same feature breakdown either way; the tool doesn't care where the text came from, only how it reads.

Does punctuation really affect the score that much?

It's one part of a weighted total, not the whole thing. Counted habits such as sentence length and contractions carry some of the heaviest weights in the method, and punctuation such as commas and semicolons sits close behind, so a piece can match the tone and still drift on them.

What happens to what I paste?

Nothing's stored beyond the scores and the differences, and that only lasts 30 days, mainly so "email me the report" has something to send. No signup, no accounts, nothing kept longer than that window.

Why does the same piece sometimes score differently against different baselines?

Because the score measures a relationship, not a fixed property of the checked piece. Change the baseline and you change what's being measured against. A piece can read close to one baseline and drift from another; that's expected, not a bug.