The skill behind this guide: Charlotte - Ghostwriter AI Skill. Holds your positions file and your samples between sessions, so the draft carries your judgements and not just your sentence length - $19, one payment, yours permanently.
View the Charlotte skill →Voice is not style. Give a model three of your articles and it will copy your sentence length, your punctuation habits and your favourite constructions, and hand you something that reads like a competent impersonation and still is not you. What identifies a person in writing is not their rhythm. It is their judgements - what they think is obvious, what they think is overrated, what they refuse to claim. A model matching your style will happily write, in your exact cadence, a sentence you would never sign.
The thing that gives a ghostwritten piece away
It is never the prose. Fluent prose is the one thing these tools are reliably good at. What gives it away is a take that is not yours: a claim slightly bolder than you would make, a concession you would not have offered, an enthusiasm you do not have, a fight you would not have picked. Readers who know you do not think "this is AI". They think "that does not sound like something he would say", which is worse, because it is a judgement about you rather than about the tool.
So the artefact worth building is not a style sheet. It is a positions file: what you actually believe, what you have said publicly before, what you concede, what you refuse to claim, and who you are arguing with. Style can be copied from three samples in a minute. Positions cannot be inferred from them at all, and that is exactly the gap the model fills in with the internet's consensus opinion.
It will supply your memories
This is the failure specific to writing in a personal voice, and it is more serious than anything in ordinary AI writing.
Personal writing runs on anecdote. Ask for a post about a lesson you learned and the model will give you a lesson, wrapped in a story: the customer call in year two, the hire that did not work out, the moment on the flight home. Well-shaped, emotionally plausible, and entirely invented. It is not lying so much as completing a pattern, because a post of that kind has a story in it and you did not supply one.
A fabricated statistic in marketing copy is bad and correctable - there is a source, and it is either right or wrong. A fabricated memory published under your name is a different object. There is nothing to check it against except you, it is the kind of detail people repeat back to you, and the correction is not "I cited the wrong figure" but "that did not happen to me".
So the rule is absolute: the model supplies no anecdote, no number, no quotation and no fact about your life. Where the draft needs one it writes a placeholder and stops. Every drafting prompt below carries that rule, and the single best way to avoid the problem entirely is the transcript method in Prompt 2.
You are a bad judge of your own voice
Read a draft of yourself written by someone else and you will accept almost anything fluent, especially if it is mildly flattering. "Does this sound like me?" is not a test, because the answer is filtered through how you would like to sound.
A usable test names a person. Pick someone who has read a lot of your writing or heard you talk for years - a co-founder, an old colleague, your sister - and ask: would they stop at any sentence in this? That question has an actual answer, and a model can attempt it if you describe the person honestly. It is the closest thing to the instinct a human ghostwriter develops after three months with a client.
Prompt 1 - the positions file, which is not a style guide
Prompt 2 - the transcript method, which solves the memory problem
The highest-value workflow on this page, and the least used. Do not prompt: talk. Record five minutes of voice notes saying what you actually think, badly, with the real examples in it. Transcribe it. Then hand the transcript over. Every fact, memory and opinion in the finished piece now came from you, and the model is doing the job a ghostwriter is actually for: structure and sentences, not substance.
Holds the positions file and your samples between sessions, works from a transcript rather than a topic, and refuses to supply an anecdote you did not give it. The persistence is the point: a positions file gets more accurate every time you correct it, and a fresh chat starts again from the internet's opinions about your subject.
View Charlotte - Ghostwriter AI Skill →Prompt 3 - drafting from a brief, when you have no transcript
Prompt 4 - surface voice, deliberately last
Match the rhythm only once the substance is right. Doing it first produces a convincing impersonation of someone saying things you do not think, which is the worst possible output.
Prompt 5 - the person who knows you
Prompt 6 - the sign-off
The last gate before anything goes out under your name.
Where the ghostwriting line actually is
Ghostwriting has a century of legitimate practice behind it. Speeches, memoirs, columns and executive posts have long been written by someone other than the person named, and the standard has always been the same: the substance has to genuinely belong to the person whose name is on it. Bring your own ideas, examples and opinions and have a tool help you write them down, and you are doing what a ghostwriter has always done.
It stops being that in two situations, and they are worth naming plainly rather than waving away.
Where the writing is the assessment. Coursework, examinations, application essays, professional or licensing portfolios, anything you sign as your own unaided work. There the point of the writing is to evidence what you personally can do, so outsourcing it is not ghostwriting, it is misrepresentation, and institutions treat it as misconduct.
Where the venue has a disclosure rule. These exist and they vary, so the answer comes from the venue rather than from a chat window.
Disclosure norms are real, and a model is not the place to ask about them
Medical and scientific journals following the ICMJE recommendations require authors, at submission, to disclose whether they used AI-assisted technologies and to describe how. Chatbots are not listed as authors, on the reasoning that they "cannot be responsible for the accuracy, integrity, and originality of the work", and humans remain responsible for anything submitted that involved them. The recommendations also warn that AI "can generate authoritative-sounding output that can be incorrect, incomplete, or biased".
That is one field's rule, not a universal one. Publishers, employers, awards and platforms each set their own and change them. Check the specific one you are submitting to, and do not ask the model what its policy is - that is precisely the kind of question it will answer confidently and out of date. Source: ICMJE.
Where it fails
| Failure | What happens | What to do about it |
|---|---|---|
| Invented memories | A plausible story about your own life, published under your name, with nothing to check it against | Forbid all anecdote. Placeholders only. Better still, work from a transcript |
| Style without substance | Your exact cadence carrying opinions that are the field's consensus rather than yours | Build the positions file first. Match style last |
| Confidence creep | Your "usually, in my experience" comes back as "always" | Record claim strength in the positions file and hold the draft to it |
| No concession | A piece that grants the other side nothing, because that is how generic thought-leadership reads | Require the concession explicitly. It is what makes a person sound like one |
| Flattery | Asked whether it sounds like you, it says yes | Ask as a named person who knows you, and ask for quotes rather than a verdict |
| Upgrade to impressive | Your plain words rewritten into more impressive ones you would never use | Instruct it to keep your phrasing where it works, and say so in the transcript prompt |
| Drift | By the third piece in a session it has slid back to the house style of the internet | Re-paste the positions file, or keep it in a skill that holds it for you |
The writing jobs that sit around a ghostwritten piece - long-form structure, headlines, scripts, publishing workflow - as ready prompts rather than a persona you load. Useful when the next piece does not need to carry a personal voice at all.
View Creative & Publishing Prompt Library →How do you install the Charlotte skill?
The download is a ZIP with SKILL.md at the root of the archive, not inside a nested folder - that folder structure is the usual reason an upload fails. In the Claude desktop app, open Customize → Skills, upload the ZIP and toggle it on.
Skills need code execution enabled, under Settings → Capabilities. Anthropic's help centre currently lists Skills on Free, Pro, Max, Team and Enterprise, while its Academy tutorial lists Pro, Max, Team and Enterprise - so if you are on the free plan, check Settings → Capabilities for your own account rather than taking either page's word for it. The full walkthrough is in the skill installation guide.
In ChatGPT or Gemini there is no upload step: open SKILL.md, copy the contents, and paste them into custom instructions. You lose automatic triggering and keep the method.
Who is this for?
Founders and executives who need to publish in their own voice and have no time to draft, consultants building a body of public writing, and anyone writing a personal essay, a newsletter or a bio where the words have to carry a person rather than a company. It works in Claude, ChatGPT or any AI chat.
The catalogue splits the writing jobs rather than bundling them, each $19 and a one-time download:
- Patrick - Newsletter Writer - the same voice problem on a weekly schedule
- Cleo - Blog Content Writer - long form, where structure carries more than voice
- Olivia - Copywriter - when the piece has to sell rather than sound like you
- Rachel - Content Strategist - deciding what to write before writing it
- Maya - Social Media Content Writer - the short-form end of a personal voice
- Max - Scriptwriter - when the voice has to be spoken aloud
The wider set is the creative, content & design skills and the content writer skills.
In summary:
Build a positions file before a style guide, because what gives a ghostwritten piece away is a judgement that is not yours, never the prose. Forbid the model from supplying any anecdote, number or fact about your life, and where you can, work from a five-minute transcript of yourself talking rather than from a topic, so every memory in the piece is one you actually have. Match your surface voice last, not first. Test the draft by naming someone who knows you and asking which sentence they would stop at. And before anything goes out, list every claim with its source, because anything sourced to the model is a problem. For a positions file and samples held between sessions, Charlotte - Ghostwriter AI Skill ($19); for the surrounding writing jobs, the Creative & Publishing Prompt Library ($9). Both work in Claude, ChatGPT and any AI chat, with a 30-day money-back guarantee.
Charlotte - Ghostwriter AI Skill
Holds your positions file and your samples between sessions, drafts from a transcript rather than a topic, refuses to invent a memory, and matches your surface voice only once the substance is right. No subscription. Yours permanently.
KissMySkills is a marketplace of 1,900+ AI skills, 60+ prompt packs, 90+ agents & free tools for Claude, ChatGPT & any AI chat.
Related Skill Guides
- How to Use Claude as a Copywriter: The Olivia Skill Guide
- How to Use Claude to Write Blog Content: The Cleo Skill Guide
- How to Use Claude as a Content Strategist: The Rachel Skill Guide
- How to Use Claude for Content Marketing: The Lena Skill Guide
- How to Use Claude as a Scriptwriter: Beats, Dialogue, Format
- How to Install a Claude Skill (Step by Step, 2026)
