AI brand guidelines: how to write rules an AI will actually follow

Most brand guidelines are written with people in mind. They use adjectives, mood boards, inspiration images, and language like "approachable but authoritative" or "warm and professional." A person reads that and, through experience and judgment, translates it into something usable. An AI cannot. It takes the description at face value and produces the average of everything that matches it, which is a different thing every time.

If you want an AI to follow your brand rules, you need to write them differently.

Why adjectives fail AI tools

"Warm," "confident," "modern," "playful," these words describe a feeling. They do not describe a rule. A person with branding experience reads "confident" and has a reference bank of examples to draw from. An AI reads "confident" and produces confident-adjacent content based on what the training data associates with that word, which could be anything from corporate keynote to aggressive startup to podcast host.

The same brief produces different outputs every time, because "confident" is not a constraint. It is a target with no coordinates.

Adjectives are the starting point for a human conversation about voice. They are not instructions an AI can execute. The fix is to turn the adjective into a rule.

Rules, not descriptions

The shift is from describing what the voice feels like to specifying what it does.

Not: "Our tone is warm and conversational." Instead: "Write in second person. Use short sentences. Do not use industry jargon. When explaining something, explain the reason, not just the instruction."

Not: "We are bold and direct." Instead: "Lead with the claim. Do not hedge with 'perhaps' or 'might.' Do not use qualifiers like 'quite' or 'rather.' One idea per sentence."

Not: "We avoid corporate language." Instead: "Do not use: leverage, utilise, synergy, robust, scalable, solution. When you catch yourself reaching for one of these, find the plain word instead."

The rule form tells the AI what to do. The description form tells it how to feel. Only one of those translates into consistent output.

Structure for machine readability

A PDF with embedded images and flowing prose is not a format an AI can parse reliably. If you are building brand guidelines to use with AI tools, structure them in plain text or Markdown.

Use clear headings so the AI can find the section it needs. Use lists for rules rather than paragraphs. Separate voice from visual from messaging, so you can include the relevant section and leave out the rest when the job calls for it.

A voice section in a machine-readable brand file might look like this:

## Voice rules

- Second person. Write "you" not "users" or "customers."
- Short sentences. One idea per sentence. If you need a comma, consider whether it should be a full stop.
- No hedging. Drop "might," "perhaps," "could potentially." If the claim is true, state it.
- Plain words. Replace: leverage > use, utilise > use, facilitate > help, robust > strong (or just cut it).
- Sentence case everywhere. No title case in body copy.

## Words we use
[list]

## Words we never use
[list]

## Example: on-brand
[short example]

## Example: off-brand
[short example with annotation]

That is a brief an AI can follow. The PDF version is a brief a person can follow.

Concrete examples are not optional

The most valuable thing you can add to AI-facing brand guidelines is examples. Not inspiration. Not reference brands. Your own examples, in your own voice, that show the rule in action.

A rule says what to do. An example shows what done looks like. When both are present, the AI has two signals pointing in the same direction and the output narrows significantly toward what you actually want.

Write at least one on-brand example and one off-brand example for each major section of voice. Annotate the off-brand one: "This is off-brand because it hedges. The claim is buried behind 'perhaps' and the sentence runs too long." An annotation teaches the AI what to watch for, not just what to copy.

Explicit dos and don'ts

A dos and don'ts list is the simplest, most reliably followed format for an AI. Keep it specific.

Not: "Do use a friendly tone." That is an adjective. Instead: "Do address the reader as 'you.' Do start sentences with a verb when giving instructions. Don't open with 'As a [brand], we...' Don't use em-dashes. Don't use exclamation marks."

The more specific the rule, the less room for interpretation. An AI given room to interpret will interpret in the direction of average. Specific rules close that gap.

Specificity matters more for AI than for people

A person reading brand guidelines brings context, experience, and the ability to ask a question. If "warm" is ambiguous, a person can look at existing work and calibrate. An AI has none of that without explicit context. It cannot ask a clarifying question. It cannot look at your past output and infer the pattern unless you show it that output directly.

This means the standard for specificity in AI-facing brand guidelines is higher than in people-facing ones. A human guideline can leave some things to judgment. An AI guideline cannot, because the AI will always exercise judgment in the direction of average when you leave it space to do so.

Write tighter rules than you think you need. You can always loosen them later. You cannot claw back the generic output that went out because the brief was vague.

Keep it current

An out-of-date brand file is worse than no file. An AI briefed from a stale document will faithfully reproduce the old voice, the old messaging, and the old rules, while you think it is following the current brand. Every time the brand evolves, the file needs to update with it.

If you are maintaining this file by hand, build in a review step whenever the brand changes. If your brand is in Brand Atlas, ATLAS.md regenerates from the live brand, so the AI is always briefed from the current version, not a snapshot from last year.

The best AI brand guidelines are not a document. They are a living brief. Write them in a format an AI can read, update them when the brand moves, and give them to the tool every time. See how ATLAS.md keeps the brief current.

Your brand is worth protecting, and that is true every time an AI tool is the one doing the work. The brief is where that protection starts.

---