GhostWriter — live demo

1. Prospect writing sample (target voice)

Paste 150+ words the prospect actually wrote. The box is pre-filled with sample data embedded in this file, not a real person. Replace it with your own text and everything below recomputes.

2. Your draft cold email

Pre-filled with sample data: a deliberately stiff draft, so the transform is visible. Replace it with your own draft.

3. Before / after against the measured targets

Metric	Sample target	Draft before	Rewritten	Delta vs target

Delta is rewritten minus measured target; a green delta is inside tolerance. Sentence length spread (standard deviation) is a diagnostic, not a target: the mechanical passes never drive it, it only moves as a side effect of splitting and merging.

4. Rewritten draft

5. Sentence length distribution

Real histogram: sentence lengths in words, counted from the sample (green) and from the rewritten draft (amber).

The rule engine, and its scope

Deterministic by design — no model, no server: the rewrite is a fixed, inspectable rule engine of contraction swaps, formality lexicon swaps, sentence splitting and merging, and canned lead-in insertion or removal. Every change is a rule you can trace, so it never invents an argument, a fact, an example or a question.

Question rate (percent of sentences ending in ?) is measured and shown, but the rewrite does not drive it. If the draft has no questions, the gap is reported and stays.

First-person matching is approximate: it adds or removes canned lead-ins such as "I think" and "I noticed", not a semantic rewrite.

Opener matching uses a small swap table of transitional adverbs, so opener vocabulary overlap can remain partial.

Contraction and formality counts are literal string matches over the whole text, so a formal word inside a quoted example is still counted.

The send gate checks literal banned phrases only. It is not a grammar, legal, or compliance review.

Metric definitions

Sentence: text split on . ! ? boundaries; sentence length is word tokens.

Spread: population standard deviation of sentence length.

Contraction rate: tokens ending in n't or in the built-in contraction list, per 100 words.

First-person rate: i, me, my, mine, myself, we, us, our, ours, ourselves, per 100 words.

Question rate: percent of sentences whose last character is ?.

Formality rate: hits from a built-in lexicon of about 50 formal words and phrases, per 100 words.

Openers: first word of each sentence, lowercased, ranked by count.