Voice Match Ghostwriter

16 views · 0 copies · 0 likes

Asking a model to write in a specific voice usually produces a caricature. It grabs the two most visible surface features - short sentences, or a fondness for dashes - and applies them relentlessly until the result reads like an impression rather than a person. The output feels close enough to be tempting and wrong enough to be embarrassing.

The failure is one of analysis, not generation. Voice is not a single dial. It is a bundle of measurable habits: average sentence length and how much that length varies, how often the writer opens with a subordinate clause, whether they prefer concrete nouns or abstractions, how they handle uncertainty, and what they refuse to do. That last one carries the most weight and gets ignored the most. A voice is defined as much by its avoidances as its habits. Someone who never uses exclamation marks and never opens with a rhetorical question has told you a great deal.

This prompt splits the work into two enforced stages. Stage one produces a written profile of the voice with no drafting whatsoever. Stage two writes new material against that profile. The separation matters because a model asked to do both at once will skip the analysis and pattern-match from vibes. Forcing the profile into visible text also lets you correct it. If the analysis claims the author is playful when they are actually dry, you fix one line instead of rejecting draft after draft.

Sample requirements. Three to five samples of 200 words or more, all from the same context. Do not mix a conference talk transcript with legal disclaimers - you will get an averaged voice belonging to nobody. If you only have one long sample, split it and say so.

How to use the profile. Save it. The profile is the durable asset, not any individual draft. Paste it at the top of future sessions and you skip the analysis stage entirely. Teams keep one profile per author and one per publication.

An honest limitation. This reproduces style, not knowledge or judgement. It will happily write a confident sentence in your voice about something you would never claim. Every draft still needs the named author to verify the substance, which is exactly what E-E-A-T expects of published work.

Ethical use. Match a voice you are authorised to write in - your own, or a client who has hired you. Imitating a named third party to publish opinions they never held is impersonation, and no prompt makes that acceptable.

You will complete this task in TWO stages. Do not begin stage two until I reply with the word CONTINUE.

STAGE ONE - VOICE PROFILE. Analyse the writing samples below and output a profile with these exact headings. Write no draft content in this stage.

RHYTHM - average sentence length, and how much the length varies. Note whether short sentences are used for emphasis and where they tend to fall.
SYNTAX - typical sentence openings. How often subordinate clauses lead. Preference for active or passive construction.
DICTION - concrete versus abstract nouns. Register from plain to formal. Any recurring word families.
STANCE - how certainty and doubt are expressed. Does this writer hedge, assert, or ask.
STRUCTURE - how paragraphs typically open and close. Typical paragraph length.
AVOIDANCES - list at least four things this writer never does. Be specific.
TELLS - up to three habits so distinctive that their absence would be noticed.

Then stop and wait.

STAGE TWO - DRAFT. When I reply CONTINUE and give you a topic, write the requested piece so it obeys every line of the profile. Then add a section titled PROFILE COMPLIANCE listing each heading above with a one line note on how the draft honoured it. Where you deliberately broke the profile, say so and explain why.

Rules for both stages. Never invent facts, statistics, or anecdotes to fill space - use a clearly marked placeholder such as [EXAMPLE NEEDED] instead. Never mention that the text was AI assisted inside the draft itself.

SAMPLES:
[paste 3 to 5 samples of at least 200 words each, all from the same context]