Most tools ask you to describe your voice. ScriptGrain measures it instead.

Most AI writing tools start with instructions. Type in three adjectives, pick a tone from a dropdown, hope the output sounds like you. It rarely does.

ScriptGrain starts with your actual writing. Not a description of it, the thing itself. We measure the patterns already sitting in your work, turn them into a reusable voice profile, apply that profile to new drafts, then measure the result against the same profile. That is how you know it matched, rather than how you hope it did.

Not just generated. Measured.

The models got better. Their default voice did not become yours.

Frontier models write well now. That was the hard problem five years ago. Not anymore. Competent is not the same as yours: every model, left to its own devices, slides toward the same safe middle. Similar sentence shapes, the same few transitions on rotation, even pacing, an opening that plays it safe because safe is statistically likely.

You can prompt your way partway out of that. It helps. It does not fix the actual problem, which is that most of what makes your writing recognisably yours was never a decision you made on purpose:

Nobody sits down and plans that. But it is not random either. It is a pattern, it repeats every time you write, and repeating patterns can be measured.

Do not ask someone to describe their voice. Read it.

Most personalisation starts with self-reporting. Pick a tone, write a paragraph of instructions, choose three adjectives and hope the model interprets them the way you meant. That can steer a model. It cannot reveal the habits you do not know you have, because you are the one filling in the form, and you can only describe what you are aware of.

ScriptGrain builds its model of your voice from writing you have already produced. Evidence, not a personality quiz. One is an instruction. The other is a target.

From writing samples to a draft you can verify

ScriptGrain is not one generation prompt. It is a loop with a measurable target at both ends.

1. Add writing you have already done

Paste it in, upload files, reuse pieces from your library, or import published work from a URL on a paid plan. PDFs and Word documents are parsed in your browser and never reach a server. One piece is enough to start; around 3,000 words across several representative samples gives the profile firmer ground to stand on. Your samples define the voice, not what you are allowed to write about next.

2. Measure the patterns

A two-pass analysis reads each sample on its own, then synthesises the evidence into one profile spanning 45 reference points. About 45 seconds. You get the measurements and a narrative explanation of what they mean when you are actually writing.

3. Generate from a brief

Choose the profile, the content type and the subject. Sixteen content types, each with its own structure, so a newsletter is not squeezed into the shape built for a LinkedIn post. Eight of the sixteen produce an outline you approve before a word of draft is written, and outlines cost no credit.

4. Measure the result

Every draft comes back with a Voice Match score against the profile that guided it, plus a breakdown of which features lined up and which drifted. When a draft misses, the polish loop revises it toward the profile using those measured gaps, then scores it again. Measure, generate, score, improve.

A writing voice is not one score

The 45 reference points sit across eight layers. Each layer can align with your voice or drift away from it independently of the others. Every attribute in the profile is named below.

Lexical

Syntactic

Tone and register

Rhetorical

Punctuation and format

Function words

Content patterns

Quirks and cadence

Not everything about writing reduces to a number. It does not, and ScriptGrain is not pretending otherwise. But enough of voice is measurable to build a genuine target, and to check a draft against it afterwards. That is the claim. Nothing bigger. Every term is defined in the writing glossary.

The result is not another prompt

It is a persistent, inspectable writing target. You can read the narrative and inspect every measured attribute behind it. The measurements themselves are deliberately read-only: hand-editing them would detach the profile from what was actually measured in your writing, which defeats the point of measuring in the first place. The profile changes through re-analysis and refinement, not by typing a new number over an old one. Two curated lists, your preferred words and your filler phrases, stay editable on a paid plan.

One profile covers a newsletter today, a LinkedIn post tomorrow, an email next week. Subject changes. Format changes. The voice underneath does not. You can inspect three complete sample profiles before building your own.

The brief says what to write. The profile says who is writing it.

These are different jobs and they should stay different. A brief carries the subject, the facts, the audience, the purpose and the format. A profile carries your recurring patterns: syntax, rhythm, habits. Keep them separate and you stop rebuilding the same style instructions inside every prompt, and the same source material can become several formats without each version quietly drifting toward a different generic voice. ScriptGrain currently supports sixteen structured content types:

Generating in your voice is only half the problem. Did it actually work?

Plenty of tools will hand you a draft and tell you it used your tone. The harder question comes afterwards. ScriptGrain measures the finished draft against the profile that guided it and returns a Voice Match score, and the supporting analysis shows which features aligned and which moved away from the target. When the result falls short, the polish loop revises it toward the profile rather than asking you to bin the draft and try another blind regeneration.

Voice Match is not a quality score. It answers a narrower question: how closely does this draft align with this measured voice profile? It will not tell you whether the idea is true, useful, original or worth publishing. Those calls stay yours, and it is precisely because the question is narrow that the answer is useful.

What the product actually lets you do

Your voice should not be trapped in one text box. A public API and an MCP server are available on every plan, including the free one, so an assistant like Claude can build or use the same profiles, generate drafts and score output. Generating through the API still spends generation credits.

We are not trying to build every AI writing feature

ScriptGrain is built around one job: keeping writing recognisably yours when AI enters the process. That focus changes the shape of the whole thing. It starts from evidence rather than adjectives. It makes the target inspectable, a persistent artefact with 45 named reference points and a narrative, rather than something buried inside a generation step you cannot see into. It separates voice from the brief. It checks the output against the same profile that guided it. And it travels.

Other tools optimise for other jobs. A generalist AI writer solves for range and speed; a brand-voice template solves for consistency across a team that never wanted individual voices in the first place. Different products suit different people, and our comparison pages are dated, price-anchored, and say where another product may be the better choice.

Because describing a voice and measuring it are different jobs

Custom instructions are genuinely useful. They are quick, built into tools many people already use, and perfectly adequate when a few explicit preferences are all that matters: tone, language, response length, formatting rules, words to avoid. The catch is that they depend on what you know to tell the model.

Most writers cannot accurately report their own function-word distribution, sentence-length variance, clause ordering or punctuation cadence. Those are exactly the patterns ScriptGrain reads from the writing itself. The two approaches can complement each other; they are not the same depth of personalisation. Custom instructions record what you know about your writing. ScriptGrain measures what the writing reveals. There is a longer walkthrough at make ChatGPT sound like you.

Human is not one generic style

Stripping a few stock phrases, or swapping every dash for a comma, does not recover a particular person's voice. It changes a handful of visible signals a detector happens to score. ScriptGrain does offer an AI-writing check, free once you have an account, and a paid humanising workflow that shows before and after scores on each pass. Both work from stylometric signals.

We will say it plainly: detector scores are imperfect signals, not verdicts, and passing one is not the same as writing something good. The sourced numbers are in our detector accuracy reference. The stronger target is not to make writing look human. It is to make it consistent with one particular human, which is why the profile stays central: it gives a revision somewhere specific to go.

You do not have one tone. You can still have one voice.

You would not write a complaint email in the tone of a birthday card, and you would not expect a founder to write a product launch exactly like a board update. Tone shifts with the occasion. Style covers surface conventions: spelling, formatting, house rules. Voice is the deeper pattern that stays recognisable underneath both. ScriptGrain treats the three as related but distinct, so a brief can ask for something warmer or more formal without throwing away the measured habits that make the writer recognisable. There is more on this at writing voice.

We would rather show what changed

In our own published study, the same underlying model was given matched briefs with and without a ScriptGrain profile. The facts stayed fixed. The measured style changed. In the first matched example the profiled draft moved from about 17 words per sentence to about 11, doubled the share of sentences under eight words from 24% to 48%, increased contractions by roughly 70%, and replaced ten em dashes with none.

In another example the average sentence length barely moved. The difference showed up in the distribution instead, with short punchy sentences at 36% of the profiled draft against 16% of the bare output. Voice is not one average. It lives across interacting patterns.

The limitation, stated without softening: one run per arm, no editing, briefs written by the person who builds the product, and neither arm blind. This is a transparent product experiment, not an independent benchmark. Its value is that the briefs, method, outputs and limits are published, so you can judge the evidence yourself. Read Same Brief, Four Voices, or browse ScriptGrain research.

Three jobs, not nine

A matching voice is not a substitute for something worth saying

ScriptGrain does not create expertise you do not have. It cannot turn a weak brief into a strong idea, it cannot verify that an argument is true, and it cannot decide whether a piece deserves to be published. It also needs something to work from: one sample is enough to begin, but several strong, representative ones will always serve the profile better than a short or uncharacteristic fragment. If you have barely written anything yet, a description-led tool might serve you better for now.

And if you love writing every word yourself, and volume is not the constraint you are fighting, the free analysis may end up being the most useful thing ScriptGrain does for you. That is the boundary of the product.

See what your voice actually looks like

Give ScriptGrain something you have already written. We measure it across the complete profile, all 45 reference points, and show you what is there. Then you inspect the result before deciding whether you want ScriptGrain to generate a single word for you. A free account includes one complete voice profile, one full analysis, one free draft to try the voice, and API and MCP access, with no card required.

Questions

What does ScriptGrain actually do?

It analyses writing you have already produced, builds a reusable voice profile across 45 reference points, applies that profile to new drafts, and returns a Voice Match score showing how closely each draft aligned.

How is that different from asking ChatGPT to write like me?

A prompt carries the preferences you know how to describe. ScriptGrain reads recurring patterns out of the writing itself, keeps them in a profile you can inspect, and checks the finished draft with the same measurements that guided it.

What writing do I need to provide?

Paste text, upload files, reuse pieces from your library, or import published work from a URL on a paid plan. One piece is enough to begin, and around 3,000 words across several samples gives the analysis firmer evidence.

Can I see the profile before paying?

Yes. A free account includes one complete voice profile, one full analysis across all 45 reference points, one free draft to try the voice, and API and MCP access. No card is required.

What does Voice Match measure?

How closely a generated draft's stylistic features align with the selected voice profile. It does not judge whether the idea is true, original, useful or worth publishing.

Is there an API or an MCP server?

Yes, on every plan including the free one, so an assistant like Claude can build or use the same profiles, generate drafts and score output. Generating through the API still spends generation credits.

Is ScriptGrain an AI detector or a humaniser?

Both exist, but neither is the central promise. The AI-writing check is free once you have an account and treats its result as a signal rather than a verdict. The humanising workflow is paid and shows before and after scores on each pass.

Is ScriptGrain always the better choice?

No. A general chatbot may be better for instant drafting with no samples to hand, and other products are stronger at other jobs. ScriptGrain is built for measuring, preserving and checking one specific writing voice.

Not ready for an account? Run the free writing style analysis, which counts twelve features in your browser and needs no signup. Weighing ScriptGrain against something else? See the honest comparisons, or check what each plan includes on pricing.