Brand Infrastructure · Template
Brand guidelines template for AI-generated content
A free, copyable brand guidelines template written as rules rather than adjectives, so the same document works for the people writing your content, the AI tools generating it, and whoever reviews it before it ships.
Why most brand guidelines stop working once AI writes
Classic brand guidelines are written for a human reader who fills the gaps with judgment. “Friendly but premium” works when a senior writer knows what you mean. A model given the same phrase has no rule to test a word choice against, so every session makes its own interpretation, and across hundreds of outputs those interpretations drift apart.
The fix is not a longer guide. It is the same brand decisions written down so they can be followed and checked: colours as exact values, voice as instructions with a test, terminology as lists, and claims as an allow-list with an owner. This template is laid out that way.
What the template covers
- Who we are: the one-sentence position every piece should be consistent with, and the facts you are allowed to cite.
- Who we write for: the reader, what they already know, and the jargon never to assume.
- Voice: three or four traits, each with a do, a don't and a pass/fail test, so “confident” means the same thing to a writer, a model and a reviewer.
- Tone: how the voice shifts by channel, with length limits, so a support reply and a launch post can differ on purpose.
- Words: exact product names and casing, say/don’t-say pairs, banned words, and how to refer to competitors.
- Claims and commitments: what anyone may say, what needs sign-off and from whom, and what no one may say.
- Grammar and formatting: headings, numbers, dates, lists, emoji and spelling: small rules that drift first.
- Visual identity: logo files and misuse, colours as exact values and tokens, type, and approved versus rejected imagery.
- Examples: labelled on-brand and off-brand pairs, with the rule each one follows or breaks.
- Review checklist: the checks every piece passes before it ships, each pointing back to a section.
- Changes: a version number, an owner and a changelog, so everyone knows which rules are current.
The template
Markdown, so it pastes cleanly into a doc, a wiki, a repo or an AI tool. Replace everything in angle brackets.
# Brand guidelines: <Brand name>
Version: <1.0> Owner: <name, role> Last changed: <date>
Applies to: <every channel / list the products, regions and teams>
## 1. Who we are
- One sentence: <what we do, for whom, and why it matters to them>
- We are not: <the category or company people confuse us with>
- Proof we may cite: <approved facts, each with its source>
## 2. Who we write for
- Primary reader: <role, seniority, what they already know>
- What they want from us: <the job they are trying to get done>
- Never assume they know: <jargon, acronyms, internal names>
## 3. Voice (the same everywhere)
For each trait, a rule a writer or a model can follow, and a test a
reviewer can apply.
- <Trait>, not <its excess>
Do: <concrete instruction, e.g. "lead with the outcome in the first sentence">
Don't: <concrete ban, e.g. "no exclamation marks in body copy">
Test: <pass/fail question, e.g. "Is the first sentence the outcome?">
- <Trait>, not <its excess>
Do: <...> Don't: <...> Test: <...>
- <Trait>, not <its excess>
Do: <...> Don't: <...> Test: <...>
## 4. Tone (what changes by context)
| Context | Tone shift | Length / format |
|--------------------|-----------------------------|---------------------|
| <Product UI> | <plainer, shorter> | <max words> |
| <Support reply> | <warmer, first person> | <max sentences> |
| <Social> | <...> | <...> |
| <Sales email> | <...> | <...> |
## 5. Words
- Our names, exactly as written: <Product Name>, <Feature name> (never <wrong forms>)
- Say / don't say:
<customers> not <users>
<...> not <...>
- Banned words and phrases: <list>
- Competitors: <name them / never name them / how to refer to them>
## 6. Claims and commitments
- Allowed without review: <claims with their source>
- Needs sign-off from <role>: <pricing, discounts, timelines, guarantees,
security or compliance statements, customer names, numbers>
- Never: <promises, comparisons or legal statements no one may make>
## 7. Grammar and formatting
- Headings: <sentence case / title case>, <max words>
- Numbers: <numerals from 10 / always numerals>; dates: <format>
- Lists: <Oxford comma yes/no>; <full stops on list items yes/no>
- Emoji: <never / only in <channels>>
- Spelling: <US / UK English>
## 8. Visual identity
- Logo files: <link to source files>; minimum size: <px>; clear space: <rule>
- Never: <stretch, recolour, add effects, place on <backgrounds>>
- Colours (exact values):
<Primary> <#hex> <token name> use for: <...>
<Secondary> <#hex> <token name> use for: <...>
<Text> <#hex> <token name> use for: <...>
- Type: <font family>, weights <...>; headings <size/weight>; body <size/weight>
- Imagery: approved examples <links>; rejected examples <links>;
what makes one approved: <criteria>
## 9. Examples (labelled)
- On-brand: "<sentence>" Why: <which rules it follows>
- Off-brand: "<sentence>" Why: <which rule it breaks>
- Fixed: "<the off-brand sentence rewritten>"
(Add at least one pair per channel in section 4.)
## 10. Review checklist (run on every piece before it ships)
- [ ] Our names are spelled and cased exactly as in section 5
- [ ] No banned words; preferred terms used
- [ ] Every claim is on the allowed list or has sign-off (section 6)
- [ ] Each voice test in section 3 passes
- [ ] Tone and length fit the context in section 4
- [ ] Colours, type and logo match section 8 exactly
- [ ] Formatting follows section 7
## 11. Changes
| Version | Date | What changed | Approved by |
|---------|--------|------------------------|-------------|
| <1.0> | <date> | <first version> | <name> |
How to fill it in
- Start from what you already have. Pull your current brand book, style guide, messaging doc and the last few pieces you were proud of. Most of the answers exist; they are just scattered or written as adjectives.
- Turn every adjective into a rule and a test. “Clear” becomes “one idea per sentence; no sentence over 25 words” and the test “Can each sentence be read once?”. If you can’t write the test, the rule isn’t finished. For voice in more depth, use the brand voice guidelines template.
- Write the words section from real mistakes. Search recent content for misspelt product names, old feature names and phrases your team keeps correcting. Those are your say/don’t-say pairs and banned list.
- Give claims an owner. Name the person or role who signs off pricing, timelines, guarantees, security statements and customer names. AI tools will produce these confidently whether or not anyone approved them.
- Use exact values for anything visual. Hex codes or design tokens, font weights and sizes, links to logo source files, and image examples labelled approved or rejected with the reason.
- Add examples last, from your own content. Take a real off-brand sentence, say which rule it breaks, and show the fixed version. One pair per channel teaches more than a page of description.
- Run the checklist on ten recent pieces. If reviewers disagree on a check, the rule behind it is still too vague. Tighten it, bump the version and log the change.
Common mistakes
- Voice traits with no limits. “Bold” without “not reckless” gives a model permission to overclaim.
- Rules only in examples. Examples show a rule; they don’t state it. Write the rule, then add the example.
- A copy per tool. Pasting the guide into every assistant and prompt template creates versions that quietly diverge. See why prompt stuffing breaks at scale.
- No version or owner. Without them, nobody can say which rules a piece was checked against, or who decides when they conflict.
Where to keep it once it’s written
Keep one source, versioned, that every person and tool reads from, rather than copies in each prompt. A shared doc with a changelog is a start; a file in a repository adds history and review; a dedicated brand-context layer lets AI tools query the current version directly. Rainbrand is one option for that last route: it turns brand guides, a website or files into context that agents such as Claude, ChatGPT and Cursor can use over MCP. Whichever you choose, the tests are the same: machine-readable, shared, versioned and reviewed. To check how far your current output has already drifted, start with a brand consistency audit.
Questions
What should brand guidelines include?
At minimum: who you are and who you write for, your voice and how tone shifts by channel, the exact words and names to use and avoid, which claims need approval, grammar and formatting rules, your visual identity with exact colour values, labelled examples, a review checklist, and a version history. The template above has a section for each.
How long should a brand guidelines document be?
Long enough that every rule is specific, and no longer. A rule that can't be tested (“be human”) adds length without adding guidance. Most teams find the voice, words and claims sections do most of the work; keep those precise and the rest short.
Can I paste this into ChatGPT, Claude or another AI tool?
Yes, it is written so a model can follow it. But a copy pasted into each tool is a copy that stops updating, and long pasted guides lose rules over a long task. Keep one versioned source and point every tool at it instead.
How often should brand guidelines be updated?
Whenever a rule changes, not on a calendar. What matters is that the change is versioned, logged in the changes table and reaches every person and tool that uses the guidelines at the same time.
Is this a visual brand book template?
Partly. Section 8 covers logo, colour, type and imagery, but this template leans towards words: voice, terminology and claims are where AI-generated content drifts most often. If you already have a brand book for design, link it from section 8 and fill in the rest.