# Brand voice analyzer: score any page against the brand, free · ScriptGrain

> Paste or link the piece that is the brand at its best, then up to three pages or drafts, and see a voice-match score for each with the exact attributes that moved. No signup, nothing stored; the full method is published.

Canonical: https://scriptgrain.com/us/tools/brand-voice-analyser

# Free brand voice analyzer: score any page against a brand voice

*By Jack Stovell · published 2026-09-21 · updated 2026-09-20*

A brand voice analyzer is only worth using if it says what it measured. This one returns the score with every attribute that moved, against a method you can read. No black box, no vague "voice health" number pulled from nowhere. Just the deltas, the bands, and the math behind them.

Paste the brand's reference piece and the pages you want checked below. The interactive check that renders here compares them directly and shows you where the writing has wandered and where it's held steady.

## What the analyzer measures

Here's the thing: most of what makes a voice recognisable is countable. Average sentence length. How much that length varies from one sentence to the next. Contractions per 1,000 words. Commas per sentence. Exclamation marks, semicolons, parentheticals, all counted per 1,000 words or per 300, depending on how rare they are. Whether there's a question mark in the body. Ellipses. The mix of I, we and you. The balance of the, a and an. Word length, split into short, medium and long. And how many of the profile's signature words and banned terms actually show up in the text.

None of that involves a model guessing. It's code, reading the piece and counting.

The rest can't be counted, so it gets judged. A small model (Claude Haiku) reads the draft for formality on a 0 to 10 scale, confidence versus hedging on a 0 to 1 scale, metaphor rate, humor register, emotional expressiveness, opening style, argument structure, rhythm, sentence complexity, clause ordering and specificity. That's the qualitative half. Combined, the counted and the judged features get weighted and calibrated into a single score between 0 and 1.

Full detail on the weighting and the tolerances lives at [/reference/voice-measurement-framework](https://scriptgrain.com/reference/voice-measurement-framework), and the broader case for measuring voice this way at all is at [/reference/brand-voice-measurement](https://scriptgrain.com/reference/brand-voice-measurement).

## How to read the score

So what does the number actually mean? Against a profile, 0.85 to 0.95 counts as on voice. Between 0.5 and 0.8, it's partial: recognisably related but drifting on a few attributes. Under 0.45 and it's off, full stop.

Without a profile, comparing two pieces directly (which is what the free check does), the bands shift slightly. 0.80 and above means on voice. 0.55 to 0.79 is drifting. Under 0.55 is off. The free check needs at least 120 words per piece to work reliably, and it stores nothing: paste, check, done.

The number on its own tells you very little though. The deltas are where the useful information sits. Say your reference piece runs long sentences with high variance, plenty of semicolons, low exclamation use. If the page you're checking comes back short and punchy with three exclamation marks per paragraph, that's not a vague "feels off" verdict, it's an exact attribute with an exact distance from the profile. You can see which lever moved and by how much.

That distance gets converted to similarity using a tolerance for each feature. Sentence length tolerance is the larger of 8 words or 70% of the profile's own average, so a naturally verbose brand gets more room to vary than a naturally terse one. Contractions get a floor of 10 per 1,000 words or 70% of the profile's rate. Formality has a flat tolerance of 3 points on the 0 to 10 scale. Confidence versus hedging: 0.35. It's all in the method page, unit-tested, and it's the same engine scoring drafts in the app, the API, the MCP server and this free tool. Nobody gets a friendlier scoring model because they paid more.

## Checking a whole site

One page tells you one thing. A site tells you a pattern, and patterns are where drift actually lives.

Paid plans monitor pages monthly: 5 pages on Writer and Operator, 15 on Studio, 50 on Agency, with the movement emailed to you as it happens. You don't go looking for drift, it comes to you.

There's also brand mimic, which builds a profile from a public site with consent rather than from your own reference material. That costs 1 brand-mimic credit, and the allowance runs 1 a month on Writer, 3 on Operator, 10 on Studio, 50 on Agency. Useful for reverse-engineering your own site's voice when nobody wrote down what it was supposed to be in the first place. For context: as of 20 September 2026, neither Pressmaster's nor Teambench's brand voice analyzer pages publish a method, a date or an author. You're taking their score on faith. That's not a dig, just an observation about what "trust us" actually costs you.

For a single-page comparison, use the [/tools/brand-voice-consistency-checker](https://scriptgrain.com/tools/brand-voice-consistency-checker). It's the same engine, scaled down to two pieces at a time.

## What a full brand profile adds

A single reference piece gives you one sample of the voice. A full profile draws on 45 attributes pulled from several pieces, not one, which matters because nobody writes exactly the same way in every single document. The average across pieces is closer to the real thing than any one piece alone.

Generation then gets held to that profile directly, and teams get a threshold to work against rather than a gut feeling about what "sounds right." That threshold is what turns a vague style guide into something you can actually check a draft against, attribute by attribute, and something new hires can hit without three rounds of "make it sound more like us" feedback.

If you haven't seen the underlying definitions, [/glossary/brand-voice](https://scriptgrain.com/glossary/brand-voice) covers the terms this page uses. And the plan comparisons, including the monitoring and brand-mimic limits above, sit on [/pricing](https://scriptgrain.com/pricing).

## Questions

### What counts as a good voice-match score?

Against a full brand profile, 0.85 to 0.95 is on voice. 0.5 to 0.8 is partial, meaning some attributes match and others have drifted. Under 0.45 is off. Without a profile, comparing two pieces directly, the bands are 0.80 and above for on voice, 0.55 to 0.79 for drifting, and under 0.55 for off.

### Does the free check store what I paste?

No. It needs at least 120 words per piece to score reliably, but nothing is saved once the comparison runs. Paste your reference piece and the page you're checking, get the score and the deltas, and that's the end of it.

### What's the difference between counted and judged features?

Counted features come straight from code reading the text: sentence length, contractions, commas, pronoun mix, and similar. No model involved. Judged features need a small model (Claude Haiku) to read for qualities that can't be counted directly, like formality, confidence versus hedging, humor register and argument structure. Both halves feed into the same weighted score.

### How is this different from checking one page at a time?

One page shows a single snapshot. Monitoring a whole site tracks pages monthly (5 to 50 depending on plan) and emails you when something moves, so drift gets caught before it spreads across the site rather than after a reader notices.

### What does brand mimic actually build?

It builds a voice profile from a public site with consent, rather than from material you supply yourself. That's one brand-mimic credit per profile, with monthly allowances from 1 up to 50 depending on plan. Handy for reconstructing your own voice when nobody wrote it down.