AI Ghostwriting: A Workflow for Client Drafts

AI ghostwriting uses an assistant to help draft words that a client reviews and publishes under an agreed name. Start with permission to use AI, an approved voice guide, and evidence for the claims; finish with a person checking the complete draft.

The practical challenge is keeping those agreements intact across revisions. fenbs can hold the approved project instructions, editorial decisions, and next steps so a connected assistant can read them in a later session. It does not replace your writing tool or approve a client draft.

What should you agree on before drafting?

Ask the client to agree on the audience, purpose, byline, review process, and permitted use of AI. Separate permission to use an assistant from permission to send confidential source material to a particular service. An approved workflow should name the tools and the kinds of material you may use.

1. Define the deliverable: a founder letter, website article, speech, or another specific format.
2. Agree on the reader and desired outcome, including what the draft should not promise.
3. Set disclosure expectations with the client and check the destination publication or platform policy.
4. Name the person who verifies facts and the person who authorizes publication.
5. Confirm how source files, drafts, and approval records will be stored and shared.

For example, a hypothetical consultant might allow AI help with outlines but require personal rewriting of the opening story. Record that boundary before generating prose. The assistant should not infer broader permission because the client approved an earlier paragraph.

Keep this workflow separate from drafting emails with AI or choosing an AI email client, which involve different sending and inbox decisions.

How do you capture a voice without inventing a persona?

Use writing samples the client has approved for this purpose. Describe observable choices: sentence length, level of formality, preferred terminology, use of examples, and how directly the client addresses readers. Avoid instructions that invite the assistant to invent personal memories or private opinions.

Ask the client to review a short sample paragraph before you draft the whole piece. A voice guide becomes useful when it predicts a choice: use a concrete opening example rather than an inspirational quotation, for instance. “Sound like me” leaves too much interpretation to the model.

• Use: plain language, short paragraphs, specific explanations, and approved expressions.
• Avoid: invented anecdotes, unsupported superlatives, and claims of personal experience.
• Keep exact: product names, technical terms, and quotations approved by the client.
• Ask first: changes to a stated opinion, sensitive topics, or a new promise to customers.

For practical writing principles, use the Digital.gov plain language series↗ (opens in a new tab), which emphasizes helping a specific audience understand the content. Keep the reader and purpose ahead of decorative style. A memorable phrase is useful only when it says what the client intends.

What evidence should the assistant receive?

Build a small source packet before requesting a draft. Include the approved brief, relevant writing samples, source links, and the exact facts the client has confirmed. Give each source a short label so a reviewer can follow the draft back to the evidence.

Separate an interview quote from your summary of the interview. A polished paraphrase should not acquire quotation marks during drafting. Likewise, a hypothetical example should stay labeled as an example through every revision.

Draft the middle section for the agreed audience.
Use only the approved brief and sources A through D.
Attach source labels to factual claims in this review draft.
List unsupported points separately instead of filling them in.
Do not invent quotes, results, or first-person experiences.

For every important statement, ask whether the source supports its full meaning. A customer describing one successful project does not establish a general performance claim. If the evidence is missing, narrow the wording or ask the client a specific question.

Use Decisions and rules in fenbs to preserve the approved evidence boundaries and voice instructions when a different assistant takes the next revision.

How should you review and disclose AI assistance?

Review in separate passes. First check the argument and evidence, then the voice, then names, numbers, quotations, and links. Reading only for fluency makes a plausible but unsupported claim easy to miss.

Send the client a clean draft and a short review note identifying unresolved questions. Keep suggested language distinct from approved language. A client comment saying “interesting idea” is not authorization to publish that idea as the client’s settled opinion.

Follow the agreed disclosure process and any applicable publication requirements. For US copyright questions, the Copyright Office report on AI and copyrightability↗ (opens in a new tab) distinguishes human creative contributions from generated material. It does not justify a blanket promise that every AI-assisted draft has the same copyright protection.

Keep records of substantive human editing and final approval. Disclosure to a client, a publication’s reader-facing policy, and a copyright registration requirement are different questions. Resolve the requirements that apply to the actual assignment rather than letting an assistant write a universal assurance.

How to do this with fenbs

1. Create a free account and add a project for the client assignment. One person can start free, with no credit card.
2. Record the client’s approved AI-use boundaries and voice requirements in Decisions and rules. Keep questions open until a person decides.
3. Add approved source references and create tasks for drafting, factual review, and client revisions on your kanban board.
4. Connect a supported assistant and ask it to read the project context before drafting. Save progress with the reviewed sections, unresolved claims, and next action before the session ends.

fenbs is not the right fit if you only need a word processor or automatic writing suggestions. It does not write inside every client document, read your chat history, or publish approved copy. Its role is keeping the work record useful between sessions.

What should the next revision start from?

Save the draft location, current version, reviewer comments, and the smallest useful next step. “Rewrite the conclusion after the client answers the warranty question” is more actionable than “continue editing.” Do not carry rejected language into a new session as if it were approved.

An AI brain is useful here because the continuing record explains the work behind the draft. For more workplace examples, see how to use AI at work and the AI at work hub. Start with one assignment whose sources and approvals you can check completely.