Start by writing your brand voice down once and storing it where every chat can use it: a ChatGPT Project or a Claude skill holding your voice rules, your price list and three or four pieces you are proud of. Then give AI real examples instead of adjectives, and edit every draft before it publishes. You can set up the first piece in about 9 minutes in ChatGPT Projects: Your Brand Voice in One Place.
Everything sounds the same when everyone types a one-line prompt into the same tool. The model fills the gaps with the average of the internet: "In today's fast-paced world", three emojis, a list of benefits any competitor could claim. The fix is not a better tool. It is context the tool cannot get anywhere else: how you actually talk, what you never say, the facts only you know, and examples of your best work.
What should a two- or three-person team set up first?
- A one-page voice guide. Who you write for, three words for how you sound, words you never use, how you sign off, and two before-and-after examples.
- One shared home for it. A ChatGPT Project or a Claude Project or skill that holds the voice guide, the current price and offer sheet, and your best examples. Everyone starts chats there.
- An examples file. Five to ten real posts, emails or pages that sound like you, grouped by type.
- A few reusable prompts with blanks for the common jobs: a launch email, a LinkedIn post, a landing page section, a customer story.
- An edit rule. Nobody publishes an AI draft without a human edit and a fact check.
Brand-focused writing platforms such as Jasper are worth a look when you produce a lot of content across many channels. For most small teams, a well-built project in the chat tool you already have is enough to start.
Why do examples work better than describing your voice?
Telling AI to be "friendly but professional" gives it almost nothing, because every brand says that. Showing it three real replies or posts gives it sentence length, word choice, how you open and close, and what you leave out. This is called few-shot prompting. Add one guard line, "copy my style, not the facts", so it does not reuse details from the examples.
What does this look like in a real lesson?
In the ChatGPT Projects lesson you set up a project for a small house-cleaning company. You choose project-only memory, paste instructions for the brand voice, the sign-off and when to use the price list, then upload the price list and the brand voice guide as sources. In a fresh chat inside the project you ask for a price on a 3-bedroom deep clean and check the price line against the list. Then, in another new chat, you ask for a three-sentence About Us, and it comes back in the brand's voice without you explaining the business again.
Give Examples: Few-Shot Prompts shows the examples step. A bike shop owner gets a generic "we" reply to a Google review, then pastes three of his real past replies between ### lines and gets one that sounds like him. For documents and designs, a brand skill that makes every doc on-brand has Claude save a florist's colors, fonts, logo rules and voice as a skill, then produce an on-brand proposal and Instagram post without being reminded.
What mistakes should I avoid?
- Publishing first drafts. AI writes fluent filler. Cut the first and last sentences and check every claim.
- Letting AI invent facts. Prices, results and dates must come from your files. Advertising claims need to be true and backed up.
- Fake reviews or testimonials. Never have AI write reviews or customer quotes. The FTC's rule on consumer reviews bans fake reviews, including AI-written ones.
- One voice file nobody updates. When an offer or price changes, replace the file in the project.
- Using AI images without checking rights. Know your tool's terms, and avoid real people's likenesses and other brands' logos.
How is this different from the marketer lesson path?
This page is about voice. The marketer start guide lays out the full sequence of lessons, the AI for Marketers post covers the starting path, and AI automation for marketing agencies covers the automations that come later. If you have to choose tools on a small budget, see Which AI tools does a two-person marketing team actually need?
Key takeaways
- Write the voice guide once and keep it in a shared project or skill.
- Show real examples, and say "copy my style, not the facts".
- Keep prices and claims in files, not in people's memory.
- A person edits and fact-checks everything before it publishes.
Frequently asked questions
- Will Google penalize AI-written marketing content?
Google says it rewards helpful, original content however it is made, and acts against content made mainly to game rankings. Content edited by someone who knows the subject, with real details, is the safe path.
- Should we tell customers we use AI?
There is no single rule for all marketing copy, but be honest where it matters: never present AI images as real customer photos or AI text as a real customer's words. Follow any platform labels for AI-generated images and video.
- How many examples should I give the AI?
Three to five good, varied examples usually work better than one or twenty. Pick pieces that differ from each other so it learns the voice, not one template.