AI can write a blog post, a product description or a support reply in seconds. Ask it to sound like your brand and it usually sounds like everybody's brand: upbeat, wordy and vaguely enthusiastic. The fix isn't avoiding AI; it's giving it, and your team, voice guidelines specific enough to follow. Here's how to write them.
Why most voice guidelines fail
"We're friendly, professional and innovative" describes almost every company. Guidelines like this fail for people and for AI alike, because they don't say what to do. Useful guidelines are specific, show examples and list the words you do and don't use.
The parts of voice guidelines that work
1. Three or four voice principles, each with "this, not that"
| Principle | This | Not that |
|---|---|---|
| Plain | "We'll call you within a day." | "We endeavour to respond expeditiously." |
| Specific | "Cut load time from 4.1s to 1.8s." | "Dramatically improved performance." |
| Warm | "Sorry, that's on us. Here's the fix." | "We apologise for any inconvenience caused." |
| Confident | "This is the right choice for most shops." | "This may potentially be suitable." |
2. Word list
- Words we use: the plain terms your customers use.
- Words we avoid: jargon, clichés and banned phrases ("revolutionary", "seamless", "unlock").
- Spelling and style: British or American English, numbers, capitalisation, how you write your product names.
3. Tone by situation
How the voice flexes: a sales page, a support reply, an error message, an apology, a social post.
4. Real examples
Before-and-after rewrites of your own copy are the most useful part, for people and for AI.
5. Facts and claims rules
What you can and can't claim, which numbers are approved, and the rule that statistics need a source.
Teaching AI your voice
- Give it the guidelines and examples every time, not just a one-line instruction.
- Show, don't tell. Three examples of approved copy beat a paragraph of adjectives.
- Ask for specifics. "Use our real numbers; don't invent statistics" prevents made-up claims.
- Set length limits. AI tends to pad; tell it the target length.
- Ban the clichés explicitly. List the phrases you never want to see.
- Review every output before it's published.
A review checklist for AI-written content
- Does it sound like us, or like anyone?
- Are all facts and numbers real and sourced?
- Has it used any banned words or clichés?
- Is it as short as it can be?
- Would a customer understand every sentence?
- Is anything promised that we can't deliver?
Our guide to whether AI content is safe for SEO covers the search side.
A one-page voice template
Fill this in and you have guidelines people and AI can use:
| Section | Your answer |
|---|---|
| Who we're talking to | |
| Three voice principles, each with "this, not that" | |
| Words we use | |
| Words we never use | |
| Spelling and style (UK/US, numbers, capitals) | |
| How we sound in support replies | |
| How we sound on sales pages | |
| How we apologise | |
| Claims we can make, with sources | |
| Three examples of copy we love |
An example brief for AI writing tools
Write a 120-word product description for our sourdough loaf. Voice: plain, warm, specific. British English. Short sentences. Use: "baked this morning", "slow-fermented for 24 hours", "Leeds". Never use: "artisanal", "elevate", "indulge", "unleash", "game-changer". Facts you may use: 24-hour fermentation; baked daily from 5am; flour from a Yorkshire mill. Don't invent any other facts or numbers. Here are two descriptions we've approved: [example 1] [example 2]
The difference between this and "write a nice description of our bread" is enormous.
AI clichés worth banning
Every brand will have its own list, but these phrases mark copy as generic, and AI tools reach for them constantly:
- "In today's fast-paced world"
- "Unlock", "unleash", "elevate", "supercharge"
- "Seamless", "cutting-edge", "game-changer", "revolutionary"
- "Look no further"
- "It's not just X, it's Y"
- "Whether you're a … or a …"
- Ending every paragraph with a rhetorical question
Banning them explicitly in your prompts and guidelines makes a noticeable difference.
Who owns the voice?
Name one person responsible for the guidelines: usually someone in marketing or the founder. They approve changes, settle disagreements and review a sample of published copy each month. Without an owner, guidelines drift and nobody notices until the brand sounds like everyone else.
Keep the guidelines alive
Voice drifts. Update your guidelines when you approve or reject copy for a reason, so the reason is recorded. Over time your examples become a library of decisions, which is exactly what makes AI tools more accurate.
Voice is part of the brand system
Voice sits alongside your visual identity; see why consistent visual design is a revenue lever and building a brand style guide. Together they make every touchpoint recognisably yours.
How Zelvorm keeps AI on-brand
Every Zelvorm project has a Brand Brain: a living memory of the client's guidelines, tone of voice, taste profile, decisions and feedback. Every AI draft is checked against it before anyone sees it, and a person signs off, so AI speeds the work up without making it sound generic.
The short version
Write voice guidelines that are specific: principles with "this, not that" examples, a word list, tone by situation, real before-and-after rewrites and rules for claims. Give them to AI every time, review every output, and keep updating the guidelines as you make decisions.
