Teach AI Your Brand Voice Before It Writes for You

This summer I noticed something about the emails AI was helping me write. They were clear and polite, and they didn’t sound like me. Clients who have known me for years would have noticed.

So I had Claude read 40 of my sent emails and compare the ones I had typed myself with the ones AI had helped draft. Some of the differences were obvious once I saw them:

• Em dashes showed up only in the AI-assisted emails. I have never typed one.
• “Hi Dan,” openers were also AI-only. I start with the name and go straight to the point: “Dan, sorry for the delay.”
• I write shorter the closer I am to someone. A note to my crew can be one line with no greeting. A client email gets more reasoning. AI wrote everyone the same medium-length, friendly email.

Those findings became a written voice guide, and every draft now goes through it.

If your team is using AI for posts, emails or web copy, this is the first thing I’d set up. It takes an afternoon.

1) Collect 20-40 real examples

Pull writing from the person or brand whose voice you want, written by hand rather than polished by committee. Sent emails are better than published copy because they’re unguarded. Get a mix of audiences: customers, partners, your own team.

2) Ask the AI to find the patterns

Paste the examples in and ask for the patterns: how messages open and close, sentence length, phrases that repeat, what gets bolded or listed. Then ask the more useful question: what shows up in AI drafts that never shows up in these? That second list is where most of the machine tells are.

3) Write a “never” list

Ours is short and specific. It bans em dashes, sales-speak like “circle back,” “touch base” and “game-changer,” cold-email openers like “Hope this finds you well,” and empty superlatives like “amazing” and “incredible.” Every brand’s list will be different. The point is that it’s written down, so nobody has to catch these by feel.

4) Describe how the voice changes by audience

Your brand doesn’t talk to a new prospect the way it talks to a ten-year customer, and AI won’t know that unless you tell it. Write two or three plain lines per audience: how long, how formal, how it opens and closes.

5) Put it where the AI reads it every time

A voice guide in a shared drive folder won’t get used. Put it in the place your AI tool reads before every draft: custom instructions in ChatGPT, a project in Claude, or the system prompt of whatever tool your team uses. If people have to remember to paste it in, they won’t.

6) Fix the guide, not just the draft

When you correct an AI draft, ask why it was wrong, and add that to the guide with the date. Ours started from my edits to one recruiting email. Each fix now helps every future draft instead of just the one in front of you.

7) Read it out loud

Our guide ends with a short test for every draft: Would I actually type this sentence? Is there a clever line to cut? Does the reader have room to push back? Reading a draft aloud catches most of what a checklist misses.

One rule that matters more than style

A voice guide controls how things are said, not what is true. Ours says plainly: never invent a number, client or credential. AI will happily fill a gap with something that sounds right. The facts still have to come from you.

If you’d like help building a voice guide like this for your brand and putting it into your team’s tools, that’s part of what our monthly retainers cover now. I’m happy to schedule a 15-minute call: book a time here.

If your team has found a different way to keep AI on-brand, I’d like to hear how you did it.