Brand Guidelines for AI: Writing a brand.md Agents Can Follow

What are brand guidelines for AI?

Brand guidelines for AI are your brand rules rewritten so a language model or coding agent can apply them without a designer in the room. They live in a plain-text file, often called brand.md, that sits next to your code or inside your AI tool's project settings. The file covers the same ground as a normal guide, but states each rule explicitly.

Your existing brand guidelines are still the source of truth. They show people what the brand looks like through layouts, mockups and examples. The brand.md translates that into instructions: which hex values to use, which words never to write, how much space the logo needs. Humans read the guide; agents read the file.

Why can't AI agents follow a PDF brand book?

AI agents struggle with PDF brand books because the most important rules in them are visual and implied. A page showing a logo on a photo with generous space around it teaches a designer the rule at a glance. An agent reading extracted text may get a caption and nothing else, and it can't infer the taste behind the layout.

Brand books also rely on judgment words such as bold, warm, confident or premium. A person who has seen the brand knows what those mean in practice. A model fills them in with its own average idea of bold, which is how teams end up with generic copy and off-brand components. Our guide to keeping brand voice consistent with AI writing tools shows the same problem from the writing side: adjectives don't steer a model, concrete rules and examples do.

What goes in a brand.md file?

A brand.md file needs every rule an agent would otherwise guess at, grouped so the agent can find the relevant section for its task. Writing agents mostly need the voice and vocabulary sections. Coding agents mostly need tokens, type, components and logo rules. Both need the facts about the company they must never invent.

If your brand also has a design system, the brand.md should point to it rather than repeat it. The split we describe in brand guidelines vs design system applies here too: the brand file says what the brand is and why, the design system and its tokens say how the interface is built.

What does a brand.md template look like?

A good brand.md template is short enough to load into every session and specific enough that two agents produce the same result. The outline below is a starting point. Copy it into a plain-text file, make each label a Markdown heading, replace the braces with your own details and delete any section you don't need.

Our free brand guidelines template covers the human-facing guide these rules come from. Start there if your brand hasn't been written down yet, then build the brand.md from it.

How do you write voice rules an AI writing agent can follow?

Write voice rules as behaviors with an example, not as personality words. "Confident" means nothing to a model; "state the recommendation in the first sentence and don't hedge with 'might' or 'could'" does. Each trait should pair what to do with what to avoid, so the model knows where the line sits.

• Turn adjectives into instructions. "Friendly" becomes "use contractions and address the reader as you".
• Ban specific words. List the stock phrases your team keeps deleting, and add to the list whenever one slips through.
• Include real samples. Two short passages from different channels steer a model better than any description.
• Add an off-brand example. One passage that sounds wrong, with a line on why, sharpens the boundary.
• State the fact rules. Tell the agent which numbers and claims it may use, and that it must never invent the rest.

Keep the voice section to a page or less. Long voice sections get diluted: the model gives every rule equal weight, and the ones that matter most get lost.

How do you write visual rules a coding agent can follow?

Write visual rules as exact values and named tokens, because coding agents can only reproduce what they can read. "Use our brand blue" invites a guess; "use --color-accent, which is #2457FF, for primary actions only" doesn't. Where your codebase already has tokens, reference them by name and tell the agent never to hard-code a raw value.

Contrast rules are worth stating outright, since agents happily put light grey text on white. WCAG 2.2 sets the AA minimum contrast ratio at 4.5:1 for normal text and 3:1 for large text, and you can link the agent to the W3C's WCAG 2.2 recommendation or simply list which token pairs pass. For the logo, point to the exact files to use; our guide to logo file formats explains why SVG is the right default for the web.

Where should brand.md live so agents actually read it?

Put brand.md wherever your agents already load context, so nobody has to remember to paste it. For coding agents, that usually means the repository: many tools read an instruction file at the project root, such as AGENTS.md, CLAUDE.md for Claude Code or Cursor's project rules, and that file can point to brand.md. File names and loading rules change between versions, so check each tool's current documentation.

For writing tools, upload brand.md to the shared project, workspace or custom-instruction area your team uses, so it loads with every chat. If you host your guide publicly, as we describe in living brand guidelines on your website, you can publish a text version alongside it, but keep the agent copy in your own tools rather than relying on agents fetching it. A public brand file is not the same thing as llms.txt, which lists pages for crawlers rather than giving rules to your own agents.

How do you keep brand guidelines for AI up to date?

Keep brand guidelines for AI current by treating brand.md as part of the brand guide, not a side document. If the two drift apart, agents will keep enforcing the old palette or voice long after people have moved on. A light process is enough.

1. 1

   Assign one owner

   The person who owns the brand guide also approves changes to brand.md. Without an owner, every teammate adds rules and the file contradicts itself.

2. 2

   Change both together

   When a color, font or voice rule changes in the guide, update brand.md in the same change, and note the date at the top.

3. 3

   Log repeated misses

   When reviewers fix the same agent mistake twice, add a rule for it. That's how the file gets sharper over time.

4. 4

   Re-test after tool updates

   Run the same three test tasks, such as a hero section, a product email and a button component, after major model or tool updates.

5. 5

   Prune quarterly

   Delete rules that no longer apply. A shorter file is followed more closely.

Rolling the file out is the same job as rolling out any guideline, and our guide to rolling out brand guidelines to your team covers it. If you're building the brand itself, our fixed-price brand identity sprint settles the identity that a brand.md is written from.