Skip to content
Search
6 min read

AI Brand Voice: Drafts That Sound Like You, With Claims You Can Back Up

Meelie Gagne

TL;DR: Getting an AI brand voice right takes three things: a one-page voice guide you paste into every request, two or three real examples of your best writing, and a claims check before anything is published. The assistant drafts. You make sure every claim is true, has evidence behind it and sounds like your company.

Ask an AI assistant to "write a post about our spring sale" and you'll get something that could have come from any company. It will probably also promise something you can't deliver, like "the best service in town". This lesson shows marketers who are new to AI how to keep an AI brand voice consistent and how to catch the claims you can't back up. You don't need technical skills. An AI assistant here means the chat tool your company has approved, and a prompt is the message you type into it.

As with any work task, follow your company's AI policy first, and use the tool it approves.

Why AI drafts sound generic

An assistant writes the most likely next words based on everything it learned from. With no other instructions, it falls back on the generic phrasing it saw most often. To sound like yours, it needs to see how yours sounds. You give it that in two ways: by describing your voice and by showing examples.

Showing examples has a name. It's called few-shot prompting, and it means putting a few samples of what you want into the prompt so the assistant can copy the pattern. Telling it who to write as is called persona prompting. You'll use both.

Step 1: Write a one-page voice guide

You may already have brand guidelines. They're often long and full of logo rules. For AI, you want one page of plain instructions. Keep it in a document you can copy from.

Include:

  • Who we are, in two sentences. What you do and who for.
  • Who we write for. Your customer, in one sentence.
  • How we sound. Three or four traits, each with a short "this, not that" example from your own writing. For instance: "Warm: 'We'll sort it out today.' Avoid: 'Your satisfaction is our top priority.'"
  • Words we use and words we avoid. Your product names spelled correctly, terms you prefer, phrases you never use.
  • Rules. Reading level, sentence length, whether you use exclamation marks, how you sign off.
  • Claims we can make. A short list of facts you can prove, with where the proof lives (for example, "Family-owned since 2009: company registration"). This list matters most, and Step 4 explains why.

Telling the assistant what to avoid works well when it's specific. The article on the negative prompt pattern goes into more detail.

Step 2: Pick two or three examples

Choose short pieces your team is proud of and that performed well: a post, an email, a product description. Remove anything about real customers that they didn't agree to share publicly. Two or three examples are usually enough to start.

Step 3: Ask for the draft

The Prompt:

You write marketing copy for our company. Follow the voice guide
and match the style of the examples below.

VOICE GUIDE:
[paste your one-page guide]

EXAMPLES OF OUR WRITING:
[paste two or three examples]

TASK:
Write a social post announcing our spring opening hours.
Facts to use: [list the facts, e.g. new hours, start date].

Rules:
- Use only the facts listed above and the approved claims in the
  voice guide. Do not add any other claims, numbers or promises.
- Under 80 words.
- After the post, list every claim you made and which approved
  claim or fact it came from.

Why This Works: The voice guide and examples set the style. The facts list and the approved claims set what can be said. The last instruction makes the assistant show you where each claim came from, which turns your check into a quick comparison.

What you get back: A short post in your voice, followed by a list of its claims and their sources. If a claim on that list doesn't trace back to your facts or your approved list, delete it.

Step 4: Check every claim before you publish

This step protects your company. In the US, the Federal Trade Commission's guidance is plain: "Advertising must be truthful and non-deceptive," and "advertisers must have evidence to back up their claims" before the ad runs (FTC, Advertising FAQ's: A Guide for Small Business). That applies whether a person or an assistant wrote the words.

AI drafts can add claims to make copy sound persuasive: "trusted by thousands", "the fastest in the area", "clinically proven". The assistant may also state a number with nothing behind it. That's called a hallucination, and in marketing it can turn into a false advertising claim.

Before you publish, check:

  1. Every number. Can you show where it came from?
  2. Every comparison. "Best", "fastest", "cheapest" and "number one" need proof. If you don't have it, cut the word.
  3. Health, safety, money and environmental claims. These are held to stricter evidence standards. If you're unsure, ask someone who knows the rules for your industry.
  4. Customer quotes and reviews. Use only real ones, from real customers, with their permission. Never ask an assistant to write reviews or testimonials. The FTC's rule on fake reviews, announced in August 2024, covers reviews by people who don't exist, including AI-generated ones (FTC press release, August 14, 2024).
  5. Names and products. Spelled right, current and actually offered.

What not to paste

  • customer lists, email addresses or any personal data from your marketing tools
  • unreleased product plans, launch dates or results your company hasn't shared publicly, unless your policy says that tool is approved for them
  • login details for your social accounts, website or email platform, ever

Make it sound like a person wrote it

Even with a good voice guide, read every draft aloud once. Cut phrases that sound like an ad from nowhere in particular. Swap a general line for a specific detail only your company could write, such as a street name, a product name or something a customer actually said. If you'd be embarrassed to have a customer ask "what do you mean by that?", rewrite it.

Where to go next

Sales reps follow up with the prospects your marketing brings in. The sales track shows how they use AI for call prep and follow-ups with the same rule about claims. More guides by job are in the Applied pillar.

Frequently asked questions

How do I train AI on my brand voice?

You don't need to train a model. Write a one-page voice guide with how you sound, words you use and avoid, and the claims you can prove, then paste it with two or three examples of your best writing into each request. Keep the guide in one document so the whole team uses the same version.

Can AI write in my brand voice?

It can get close if you give it a clear voice guide and real examples. Expect to edit the first drafts, and add any phrase you keep fixing to the "words we avoid" list so the next draft is better.

Is AI-generated marketing content legal?

Using AI to draft is generally allowed, but the content must follow the same advertising rules as anything else. In the US the FTC requires ads to be truthful and backed by evidence, and its fake reviews rule bars reviews from people who don't exist, including AI-generated ones. Check with someone who knows the rules in your industry and country.

How do I stop AI from making things up in marketing copy?

Give it the facts to use and a list of approved claims, tell it not to add any others, and ask it to list every claim it made with its source. Then check each one before you publish.

Should I tell customers I used AI to write content?

Follow your company's policy and any rules for your industry or platform. Whatever you decide, the content itself must be accurate, and a person should review it before it goes out.

Rather have someone set it up for you?

That's a fair call. LTFI, the agentic business operating system from Kief Studio, can set up the work these guides describe and run it with you. No website yet? Kief Studio will build you one for free.