ScriptGrain vs Claude

ScriptGrain makes Claude better at being you. Projects and Styles apply what you can describe and paste. ScriptGrain measures your voice as 45 attributes from writing you have already published, connects to Claude over MCP so it writes from that profile, and scores every draft against the measurement.

The actual difference

A Project holds what you can say about your voice and the samples you paste; most of a voice is what you cannot say. ScriptGrain measures 45 attributes from your published writing, connects to Claude over MCP so Claude writes from the measured profile, and scores every draft against it. In SGR-002 a one-page measured prompt got GPT-5 to 0.62 against the author and the profile got to 0.83. It is not a rival to Claude; it is the measurement Claude does not have.

Projects cost nothing on any Claude plan, and Pro at $20 a month buys a better model and more usage rather than a better memory of your voice. ScriptGrain's free plan measures your voice with no card and the connector is on every plan; paid plans from £12 a month add regular generation, polishing and the monthly monitor.

ScriptGrain is not a replacement for the assistant you already use. It measures your voice from writing you have published and hands that profile to ChatGPT, Claude, Muse or any MCP client through its connector, then scores every draft against it. ScriptGrain makes your AI better at being you.

Connect ScriptGrain to Claude

Every AI voice tool compared: prices, scores and API access, checked

ScriptGrainClaude
Entry price£12/mo (Writer, 1 voice profile)Free (Projects are on every plan; up to 5 on Free)
Free tierYes, See your own voice profile and try one draft. No card.Claude's free plan includes up to 5 Projects and user preferences
Voices on entry plan1 (3 on Operator, £29/mo)A Project's instructions and knowledge files, plus a Style set from samples
How the voice is built45 measured style attributes, extracted from writing you have already publishedYou write instructions and upload files into a Project, or create a Style from writing samples, and Claude applies them to each response. It is a description plus examples, read fresh each time, with no score on the result and no record of the habits you cannot articulate.

Claude figures read from their pricing page on 20 September 2026.

What Claude is built for

Claude is a general-purpose assistant. Research, code, long documents, a half-formed email you need turned into something sendable. Voice is one thing it can do among many, and that's the honest place to start.

Projects and Styles are its personalisation layer. A Project is a knowledge base (uploaded documents, text, code) plus written instructions, scoped to the chats inside it. When the knowledge outgrows the context window, retrieval switches on automatically, so you don't manage that yourself. Free accounts get up to 5 Projects; paid plans get unlimited. Styles are narrower. You upload a writing sample and Claude infers a matching style, or you describe tone, structure and vocabulary in words, and it applies to every response until you switch it off.

To be fair, that's a lot of value for no money. Most writers already have Claude open. Setup takes minutes. You can be running a Project full of your past work before lunch. If your writing job is broad, with voice as one requirement among several, Claude is built for exactly that.

How Claude handles voice

Both features work the same way underneath. Description plus examples, re-read each time.

You tell Claude what you sound like. You paste or upload samples. Claude reads all of it again at the start of a chat and does its best impression. Neither Projects nor Styles scores the output against anything. There's no number that says "this draft is close" or "this one drifted." You read the result and judge it by ear.

That works better than it sounds, for a reason. Claude is a strong writer, and a good sample plus a clear description gets you a long way. A Style built from one of your better pieces will often catch your tone, your typical opening, the feel of your paragraphs.

The limit is what you can put into words. Ask yourself how long your sentences run, how often you reach for a contraction, whether you lean on semicolons or dashes, and how much your sentence lengths swing from one to the next. Most people can't answer. So most of a voice never makes it into the instructions, and the samples have to carry it by implication. Claude infers from them, and inference is a guess. Sometimes a good one. Nobody can check.

There's also no memory of the verdict. If you edit a draft heavily because it sounded off, that correction doesn't become a measurement. It's you, doing the comparison by hand, every time.

How ScriptGrain handles voice

ScriptGrain starts from the other end. It doesn't ask you to describe your voice. It measures it from writing you've already published.

The result is a profile of 45 attributes in 8 layers: lexical, syntactic, tone and register, rhetorical, punctuation and format, function words, content patterns, and quirks and cadence. Some of those are the things you'd never think to write down. Average sentence length and its variance. Contraction rate. Comma density. Semicolon and dash habits. Formality on a 0 to 10 scale. Confidence versus hedging. Opening style, argument structure (evidence-first, conclusion-first, narrative), paragraph openers, pronoun mix. The full set of attributes is documented, and so is the method behind the measurement.

A profile needs at least one sample of 50 words or more. Three or more pieces and about 3,000 words is the recommendation. Extraction usually takes under a minute; the median measured on 2026-09-28 was 39 seconds.

Then comes the part Claude doesn't have. Every draft is scored from 0 to 1 against the profile. That's voice match. On-voice drafts typically land between 0.85 and 0.95. Partial matches sit from 0.5 to 0.8. A clearly different voice falls under 0.45. The score arrives with per-feature deltas, so you see the draft's value next to the profile's value for each attribute. You're told where it drifted, not just that it did.

Polish takes a draft and revises it toward a target score, 0.9 by default. Drafts are written with the profile as a constraint, and the app learns from the edits you save.

The profile reaches Claude over MCP, and reaches ChatGPT, Gemini, Cursor, Meta Muse and any other MCP client the same way, plus a REST API. It's read live each time. It is never a pasted prompt. Both the connector and the API are on every plan, Free included.

The same job in each tool

Take one job. You write a fortnightly newsletter, you've published a couple of dozen issues, and you want Claude to draft the next one in your voice.

In Claude alone, you create a Project. You upload several past issues as knowledge and write instructions: conversational, dry, short paragraphs, never salesy. Maybe you build a Style from your best issue too. Then you ask for a draft on a new topic. It comes back, and it's decent. The tone is about right. But the opening is a little tidier than you'd write, the sentences are more even than yours, and there's a closing line you'd never use. You fix those by hand. Next issue, you fix the same things again, because nothing recorded that they were wrong.

With ScriptGrain connected, the job starts differently. Your published issues are measured once, and the profile says, among other things, that your sentence lengths swing hard, that you open with a direct claim, and that you use contractions constantly. Claude connects to the profile over MCP and writes from it. The draft comes back scored. Say the score lands in the partial-match band, somewhere from 0.5 to 0.8. The deltas show the gap: sentence variance too low, formality a notch high. You polish toward 0.9, or you edit and save, and the app learns from what you changed.

So the drafting model is the same in both cases. What differs is whether there's a measurement underneath it.

Be realistic about effort, too. In Claude alone, you're up and running in minutes. With ScriptGrain you add a measurement step and a connector. That's a small cost, and it buys something the first route can't give you: a number you can track from draft to draft.

Which one fits which situation

Some situations are clear-cut.

Start with the question of whether you can check your own drafts. If you can, by ear and with confidence, Projects and Styles may be plenty. If you keep finding yourself making the same corrections, that's the signal. The corrections are a voice you couldn't describe.

On cost, compare like with like. Claude Pro is $20 a month ($17 billed annually) as read on 2026-09-30, and it buys a better model and more usage, not a better memory of your voice. ScriptGrain's free plan measures your voice with one profile and one full analysis, no card needed, so you can see the profile before paying anything. Paid plans start at £12 a month with Writer. The full breakdown is on the pricing page.

Using ScriptGrain alongside Claude

The practical setup is short. Measure your voice first, from your published writing. Then, in Claude's Settings under Connectors, add ScriptGrain as a remote MCP server. claude.ai and Claude Code both act as MCP clients, so either works. There's a full walkthrough at /works-with/claude, and the MCP and API page covers the technical side.

Keep your Projects. They still do the job they're good at: holding background, briefs, reference documents, a client's product details. The profile handles voice; the Project handles context. Those are different jobs, and they don't collide. You can drop any voice instructions from the Project that the profile now covers, which tends to make the instructions shorter and clearer.

A few habits help once both are running.

  1. Score every draft that matters. The number is only useful if you look at it.
  2. Read the deltas before you edit. They tell you what to fix, which beats rereading the whole piece hunting for what feels off.
  3. Re-measure when your writing changes. If your style has moved on, a profile built from older work will pull drafts backwards.
  4. Use the brand voice consistency checker or the writing style analysis tool when you want a quick read on existing text without drafting anything.

If you want to check a draft for stock phrasing as well, the AI cliché checker covers that separately.

Moving away from Claude isn't part of this. Nothing here replaces it. The profile is read live each time, so there's no export to maintain, nothing pasted that goes stale, and if you switch assistants later, the same profile comes with you to ChatGPT, Gemini or Cursor. The comparison matrix shows how the wider set of tools stacks up, and /vs lists the other head-to-heads.

At the end of the day, the choice is smaller than it looks. Claude writes. ScriptGrain measures. One keeps doing what it does best, and the other tells you whether it sounded like you.

Where Claude is stronger

The assistant many writers already draft in. Projects hold instructions and knowledge files; Styles let you set a writing style from samples. Both apply what you give them; neither measures you.

If you want one assistant for research, code, documents and quick drafts, Claude is that product, and for replies where voice does not matter a Project is plenty. Keep Claude; give it your measured voice.

Questions

Is ScriptGrain better than Claude?

They solve the problem differently. A Project holds what you can say about your voice and the samples you paste; most of a voice is what you cannot say. ScriptGrain measures 45 attributes from your published writing, connects to Claude over MCP so Claude writes from the measured profile, and scores every draft against it. In SGR-002 a one-page measured prompt got GPT-5 to 0.62 against the author and the profile got to 0.83. It is not a rival to Claude; it is the measurement Claude does not have. If you want one assistant for research, code, documents and quick drafts, Claude is that product, and for replies where voice does not matter a Project is plenty. Keep Claude; give it your measured voice.

How much does Claude cost compared with ScriptGrain?

Claude's entry plan is Free (Projects are on every plan; up to 5 on Free) (Pro $20/mo or $17/mo billed annually; Max from $100/mo; Team $25 a seat/mo), with this free tier: Claude's free plan includes up to 5 Projects and user preferences. ScriptGrain starts at £12/month (Writer, 1 voice profile) with a free plan: See your own voice profile and try one draft. No card. Claude figures read from their pricing page on 20 September 2026.

Does Claude learn my writing voice?

You write instructions and upload files into a Project, or create a Style from writing samples, and Claude applies them to each response. It is a description plus examples, read fresh each time, with no score on the result and no record of the habits you cannot articulate. ScriptGrain instead measures 45 style attributes from writing you have already published, and scores every draft against that measured profile.

Other comparisons