# Brand voice AI tool for agencies: one measured profile per client · ScriptGrain

> One measured voice profile per client, built from the client's own writing, every draft scored against it, seats for the team, and an API and MCP server on every plan. Studio £99 a month for 10 client profiles; Agency £299 for unlimited.

Canonical: https://scriptgrain.com/for-agencies

# ScriptGrain for agencies: every client voice, measured

*By Jack Stovell · published 2026-08-03 · updated 2026-09-20*

ScriptGrain gives agencies one measured voice profile per client, not one shared prompt. Every draft gets scored against that profile before it ships. Seats let writers work inside the same client without drifting apart. The API and MCP server put all of it inside Claude, ChatGPT, Cursor or your own pipeline.

Here's the thing: you don't need to trust that a draft sounds right. Run the free [multi-client voice drift check](https://scriptgrain.com/tools/multi-client-voice-drift-check) below on one of your own client's real pieces against a second one and see what the numbers say. No account, no card, just a score.

## The problem is contamination, not talent

Your writers are good. That's not the issue. The issue is that a writer who spends Monday on a fintech client and Tuesday on a skincare brand carries something across, a rhythm, a habit, a favourite sentence shape. It's not laziness. It's how brains work. And the client can't always name what's wrong, they just feel that the piece is "off" this week.

So you end up with review cycles built on gut feeling. Someone senior reads it, frowns, says "this doesn't sound like them," and nobody can point at the sentence that broke it. That's expensive. It's also avoidable, because voice contamination is measurable. Sentence length, contraction rate, comma density, question use, all of it moves when a writer slips into their default mode instead of the client's. Score it and you find the drift before the client does.

## One measured profile per client

Each client gets their own profile, built from three or more of their own published pieces (about 3,000 words total) or extracted from their public site with consent, as a brand mimic, at one brand-mimic credit. That profile covers all 45 attributes ScriptGrain tracks: sentence rhythm, pronoun mix, article balance, formality, humour register, and the rest.

Every draft you write for that client gets scored against their profile, not against a generic "good writing" benchmark. A score of 0.85 to 0.95 means the draft is genuinely in voice. 0.5 to 0.8 is partial, usable with edits. Under 0.45 means start again. You set the threshold, agency default is 0.85, and write it into the [brand voice guidelines](https://scriptgrain.com/reference/brand-voice-guidelines-template) so the whole team is polishing to the same number.

## Onboarding a client voice in a day

You don't need weeks of workshops. Feed the tool three or more published pieces, or point it at their site with consent for a brand mimic. That's it, profile built. If you're wondering how much text the model actually needs to lock a voice down reliably, the answer is covered properly in [how many samples does a voice profile need](https://scriptgrain.com/research/how-many-samples-does-a-voice-profile-need).

Set the threshold. Write the guideline doc. Brief the writers. By end of day you've got a client onboarded with a measured target instead of a vague adjective list like "friendly but authoritative." Fair enough, adjectives have their place in a kickoff call. They don't help a writer at 4pm on a Thursday trying to hit a deadline.

## QA by number, not by gut

Here's what changes day to day: a draft comes in, it gets scored, and the features that moved get named. Not "this feels off" but "contraction rate is half the profile's and sentence variance has collapsed." That's something a writer can fix in ten minutes instead of guessing.

Polish pushes the draft toward 0.9. It costs one credit, with a free skip when the text already passes, so you're not spending credits proving something is already fine. Page rewrites cost one credit per 1,500 words. Outlines, voice-match scoring, comparing two pieces, AI-detect and reading profiles are free, always. Credits only get deducted once a draft is actually delivered, so a rejected first pass costs you nothing.

## Several writers, one client, measured

This is the bit that actually saves agencies money. Put three writers on the same client and their scores tell you something a style guide never could: the spread between them is your real measure of consistency. Tight spread, the client sounds like one person no matter who typed the sentence. Wide spread, you've got a problem before the client notices one.

Set per-member credit caps so nobody blows the monthly allowance on one panicked afternoon. Shared workspaces mean the whole team sees the same profile and the same threshold, not five interpretations of "sounds like them." And the monthly monitor scores the client's live published pages on its own, 15 on Studio, 50 on Agency, and emails you the drift. You find out their voice has moved before a client email does.

## Wired into the tools you already use

The profile lives on the API and MCP server on every plan, read live from wherever you're already working. Claude, ChatGPT, Cursor, or your agency's own pipeline, all of them can pull the profile and score against it mid-draft. Nobody's exporting a prompt into a doc and hoping it survives copy-paste. The [MCP tool reference](https://scriptgrain.com/reference/mcp-tool-reference) covers exactly what's callable and how.

That matters for agencies specifically because your writers aren't all in the same tool. One prefers Claude, one lives in Cursor, one's still drafting in ChatGPT. Fine. The profile doesn't care which door you come through, the score is the same wherever you check it.

## Plans for agencies

Two plans are actually built for agency work. Studio, at £99 a month ($134), gives you 10 profiles, 400 generations, 20 extractions, 10 brand mimics, 15 monitored URLs and 3 seats included, with extra seats priced on the pricing page. Agency, at £299 a month ($405), goes unlimited on profiles, 1,500 generations, unlimited extractions, 50 brand mimics, 50 monitored URLs and 10 seats included.

Writer and Operator exist too, but they're built for solo writers managing one or a few voices, not a shared client roster. If you're a solo writer or ghostwriter reading this and thinking the agency tiers sound like overkill, they probably are, have a look at [for ghostwriters](https://scriptgrain.com/for-ghostwriters) instead. Full numbers for every tier sit on [pricing](https://scriptgrain.com/pricing), and the scoring method behind all of this, including how the 0 to 1 scale gets calibrated, is written up at [the voice measurement framework](https://scriptgrain.com/reference/voice-measurement-framework), with the field compared at [brand voice tools for agencies](https://scriptgrain.com/reference/brand-voice-tools-for-agencies).

## Questions

### What counts as voice drift?

Drift is the gap between a piece and the profile it's meant to match, or between two pieces from the same source over time. It's measured, not felt. The full definition and how it's scored sits at [voice drift](/glossary/voice-drift). Anything under 0.45 against a profile counts as off; anything over 0.85 counts as genuinely in voice.

### Do we need a card to try it?

No. The Free plan needs no card and includes one voice profile with one full writing analysis across all 45 attributes. Seeing your profile costs nothing. Generating with it needs a paid plan, since generation draws on credits and the free tier carries none.

### How many samples does a client need before we trust the profile?

Three or more published pieces, roughly 3,000 words total, or a brand mimic pulled from their live site with consent. That's usually enough to lock the profile reliably. The reasoning behind that number, and what happens with fewer words, is in [how many samples does a voice profile need](/research/how-many-samples-does-a-voice-profile-need).

### What happens when a writer's draft scores low?

The score names which features moved: sentence length, contraction rate, comma density, whichever attributes drifted from the client's profile. Polish pushes the draft back toward 0.9 for one credit, free if it already passes. Nothing ships below your agency's threshold, 0.85 by default, without someone consciously overriding it.

### Can we run this alongside Claude or ChatGPT instead of replacing them?

Yes. The API and MCP server sit on every plan and connect straight into Claude, ChatGPT, Cursor or your own pipeline. You keep writing wherever you already write. ScriptGrain reads the client's profile live and scores the draft in place, it isn't asking you to change tools.