AI for Ghostwriters: Keep Every Client's Voice Separate

AI helps ghostwriters most with the parts around the writing: turning interview transcripts into notes, keeping each client’s voice rules handy, drafting first versions, and repurposing finished pieces.

The risk is that every client starts to sound like the same AI. The fix is to give the AI a separate, well-kept source of truth for each client and make it read that first, every time.

What can AI do for a ghostwriter?

Think of the job in stages. AI is strong in some and weak in others.

Stage	Where AI helps	Where you still lead
Intake	Summarize call transcripts, pull out stories, opinions and phrases	Asking the follow-up question that gets the real story
Voice	Check a draft against written voice rules	Hearing what makes the client sound like themselves
Outline	Propose structures from your notes	Choosing the angle the client will stand behind
Draft	Fast first drafts from an outline and source notes	The lines that carry the piece
Edit	Tighten, cut, flag repeated phrases	Final judgment on tone and claims
Repurpose	Turn one essay into posts, threads, an email	Picking which pieces are worth repurposing

The pattern: AI does well when the raw material is already on the page. It does poorly when it has to invent the client’s experience or opinions, and that’s exactly what a ghostwriter must never let it do.

Why does AI make different clients sound the same?

Because each chat starts from the model’s defaults. Say your only voice guidance is “write like Sam, keep it casual.” You’ll get the model’s idea of casual, and it’s the same for every client you have.

Three things cause most of the sameness:

• Thin voice notes. Adjectives like “warm” or “punchy” don’t constrain anything. Real examples do.
• Mixed context. One long chat with several clients’ material bleeds phrasing across them.
• No memory of edits. You fix the same AI habits in every draft because nothing records that this client never says “synergy.”

How do you capture a client’s voice for AI?

Build a voice file per client from their own words, not your description of them. A useful one has:

• A set of real sentences from interviews and past posts, picked because they sound most like the client.
• Words they use and words they avoid. “Customers,” never “users.” Says “honestly” a lot; keep one per piece.
• Opinions on record. Positions they’ve taken that drafts must stay consistent with.
• Stories they’ve told, in short form, with the details that make them true. AI should reuse these, never invent new ones.
• Formatting habits. Short paragraphs, no headers on LinkedIn, sign-off style.

Then keep it current. Every time the client edits a draft, note what they changed and why. A few rounds of real edits teach more than any adjective. The guide on making ChatGPT remember your brand voice covers the mechanics for a single brand; for several clients, the next section matters more.

How do you keep client work separate?

Separation is both a quality issue and a trust issue. Clients hire you on the understanding that their unpublished ideas stay theirs.

One workspace per client. In chat apps, that means one project per client, with only that client’s files and instructions. Never paste two clients’ material into the same conversation.

One notes collection per client. Interview notes, voice file, published pieces, ideas in progress. If you keep these in one big folder, it’s only a matter of time before a draft borrows from the wrong person. The case for strict separation is laid out in one vault per brand.

Check your AI’s data settings. Each AI provider has its own settings for chat history and whether your chats are used to improve their models, and these vary by plan. Read them for every tool you use with client material, and tell clients which tools you use if your contract calls for it.

If you write for several founders or brands at once, managing multiple brands with AI goes deeper on keeping voices and context from crossing.

What does a good AI ghostwriting workflow look like?

Here’s a workflow you can adapt. It assumes you interview the client for each piece.

1. Record and transcribe the interview. Use whatever transcription your call or recording tool offers.
2. Turn the transcript into notes, not a draft. Ask your AI to pull out each story and opinion as its own short note, filed under the client. Check them against the recording.
3. Outline from the notes. Pick the angle yourself. Have the AI propose a few structures using only those notes.
4. Draft with the voice file loaded. Tell the AI to read the client’s voice file first and to use only stories and facts from the notes.
5. Edit by hand. Rewrite the opening and the closing yourself. Cut anything the client wouldn’t say out loud.
6. Log the client’s edits. When feedback comes back, add the changes to the voice file.
7. Repurpose the approved piece. One essay can become several posts and an email; see how to repurpose content with AI.

The notes step is the easy one to skip, and it’s the one that pays off. Notes from interviews become a library. Months later, a client asks for a post about hiring, and you already have stories on it in their own words.

How do you use AI without losing the client’s trust?

A few rules worth holding to:

• Never invent experiences or quotes. If the notes don’t have it, ask the client.
• Fact-check every claim. AI can state wrong things in a confident tone.
• Be clear about your process. Agree up front whether and how you use AI, the same way you’d agree on rounds of revisions.
• Keep the final pass human. The client is paying for judgment. Make sure there’s some in every piece.

FAQ

Is it ethical for ghostwriters to use AI?

Ghostwriting has always meant someone else shapes the words. Using AI as a tool inside that is a matter of agreement with your client. Be open about your process, and never let AI invent the client’s stories or opinions.

Which AI is best for ghostwriting?

You may end up using more than one: one model for drafting, another for editing or research. What matters more is that each one reads the same client notes and voice file, so switching tools doesn’t reset the voice.

How do I stop AI from sounding generic?

Give it real sentences from the client, a list of words to avoid, and the edits the client made last time. Then rewrite the opening and closing yourself.

How much of the work can AI do?

It can speed up intake and first drafts, and it makes repurposing quick. The interview and the final voice pass are still yours.

Where Valta fits

Valta is a hosted vault of Markdown notes that Claude, ChatGPT and other AIs connect to over MCP. On Pro you get up to five vaults, so each client can have one. Each vault has a profile (_vault.md) with the client’s voice and rules, and your AI reads it before writing. Each vault also has its own connector URL that opens only that vault, so an AI connected through that URL can’t open another client’s vault.

Your AI does the writing; Valta stores the interview notes and the voice file, and suggests links between them as they’re saved. Everything exports as plain Markdown whenever you want it. Set up your first client vault.