
Voice Match: Write Like You, Measured & Proven
Write like you. Measured, not guessed.
Most AI rewrites sound like the AI. Voice Match builds a measurable profile of how you actually write, then has your agent rewrite a draft in that voice, and uses an offline script to prove two things: the draft moved closer to your own range, and the facts did not change.
How it works
- Profile: the script measures 68 surface features of your own samples: sentence length and its spread, short and long sentences, paragraph length, punctuation, contractions, I / we / you, questions, sentences that start with "And" or "But", and the rate of 40 common words
- Baseline: it measures your draft against your own range and tells you what differs in plain words, for example "your sentences average 7 words, the draft's average 23"
- Rewrite: your agent rewrites toward your habits, keeping every fact
- Check the facts: the script compares the original and the rewrite: numbers, links, email addresses, quoted text, names, yes/no words, headings and length
- Measure again: repeat until the verdict is within your range or you stop
What you get
- A verdict: within your range, borderline or outside your range, judged against how far your own samples are from each other
- The distance before and after each rewrite
- The strongest differences, in plain words
- Stock phrases ("moreover", "delve", "robust") that appear in the draft but never in your writing
- A fact-preservation report that catches a changed number, a dropped link, an altered quote or a flipped yes/no
- A warning when your samples are too few, too short, too mixed or too similar to trust
Included: 5 synthetic sample texts and a demo showing two rewrite rounds and one bad rewrite, with expected reports.
Your voice only
- Samples must be your own writing. Voice Match is built to write as you, and its instructions tell the agent to decline requests to imitate a named person.
- It is not for passing work off as someone else's, or for assignments and statements where your own authorship is required.
- It is not an AI-detector tool: it does not test, predict or evade any detector.
Where it runs and what you need
Voice Match is a download package. You unzip it and use it in your own environment. It does not run through Capafy's cloud, and no AI model is included in the price.
You provide:
- An AI agent or assistant that can read a SKILL.md file, read and write files in a working folder, and run shell commands. The instructions are written for a Claude-style agent; other agents have not been tested.
- Python 3 (standard library only, nothing to install; developed on Python 3.13, other 3.x versions not tested).
- Your own model or provider account. Any model usage fees are charged by your provider, not by this package.
- Three or more samples of your own writing, 1,500 words or more in total, ideally the same kind as the draft.
Data sharing
Nothing is sent to Capafy or to the author by this package. The script runs on your computer, makes no network calls and imports no network libraries. Your samples and drafts are read by whichever agent and model provider you use, under their terms. The profile file holds numbers, a word list, short recurring phrases and hashes, not readable sentences, but it still describes how you write, so keep it private.
Honest limits
- It measures surface features only. It does not show that a draft sounds like you to a reader, that its content is true, or that its meaning is unchanged.
- Measured on synthetic template writers, same-style drafts of about 260 words landed within range 81% of the time, borderline 16% and outside 3%, and a different style landed outside every time. Real writers are less consistent and less distinct, so expect noisier results; with similar writers the verdict is a guide, not proof.
- Needs at least 3 samples to give a verdict; 2 samples and 600 words is the minimum to run.
- English prose only. Tables, code and links are left out of the measurement.
- Not legal, medical, financial or professional advice. For adults (18+).


