Which brand guideline formats can AI tools actually use?

What a format has to carry

A brand guide gives an agent rules with their reasons, exact values such as hex codes, files such as the logo SVG and fonts, and visual references. Each format keeps only some of these intact, and the agent needs the current version. How to keep AI-generated content on brand covers what each output needs.

The formats compared

PDFs and slide decks

In their APIs, Claude and OpenAI’s models read a PDF as each page’s text plus an image of the page, and Gemini reads PDFs with vision. Three things still go wrong:

• Size. Anthropic estimates 1,500 to 3,000 tokens of text per page, so a 60-page brand book is at least 90,000 tokens, mostly irrelevant to any one task.
• Pictures, not files. The agent redraws or crops a pictured logo, and estimates from pixels any swatch printed without its value.
• Copies. Each upload stays the version you uploaded.

Decks lose more: OpenAI’s API reads only the text of .pptx and .docx files, and Gemini reads non-PDF documents as plain text. Export decks to PDF, and check your chat app, which can differ from its API.

Keep the PDF for people. For agents, pull out its rules as text and its logos and fonts as files; the Markdown brand guidelines template shows how.

Brand portals and web pages

A guidelines portal can hold the current brand, but an agent’s tools decide what it can read. Claude’s web fetch tool, for example, fetches only URLs already in the conversation, doesn’t support pages rendered with JavaScript and won’t download font files or a ZIP of logos. Pages behind a sign-in may need an authenticated browser or an integration.

Two routes help: publish a Markdown version of each page and list them in an llms.txt file, a 2024 proposal for pointing agents at clean text, or connect agents to the portal, as Frontify’s MCP server does.

Markdown and BRAND.md

Markdown is plain text with headings, tables and links that any tool can read, the plainest way to give an agent rules and values. But it describes files rather than holding them, and Standards, a brand guidelines platform, warns that a downloaded copy of its Markdown export “will not reflect changes published after the file was downloaded”.

BRAND.md, a draft spec (version 0.3.0, by Caio Pizzol), puts strategy, voice and visual layers in one file in a project’s root. Its boundary test is useful even if you never adopt it: “If it should survive a complete visual redesign, it belongs in BRAND.md.” Interface details go to a DESIGN.md, and logos appear only as a note of where they’re kept.

Rules files such as AGENTS.md, which Cursor reads, work best as a pointer to where the brand lives; a palette pasted into each repository is one more copy to update. Rainbrand’s Cursor and Claude Code setup pages compare a rules file with a connection to a maintained brand.

Agent Skills (SKILL.md)

An Agent Skill is a folder with a SKILL.md file at its root, plus optional references/, assets/ and scripts/ folders, in an open standard that OpenAI’s skills for ChatGPT and Codex also use. Per the specification, agents load about 100 tokens of name and description at startup, the full SKILL.md (under 5,000 tokens recommended) when a task matches, and other files only when needed, so a colour task opens only the colour file. An assets/manifest.json can describe each image’s role:

SKILL.md
references/
├── voice-and-messaging.md
├── color-and-type.md
├── logo-and-marks.md
└── imagery.md
assets/
├── manifest.json
├── logos/
├── fonts/
└── images/

The description in its frontmatter decides when the skill loads, so name the work it covers:

---
name: brand
description: Apply our brand when creating or reviewing social posts, emails, web pages, decks, packaging or brand copy. Use this skill whenever a task asks for branded work. Includes voice rules, colour and type values, logo files, fonts and reference images.
---

Skills are copies too: Anthropic says custom skills “do not sync across surfaces”, so one uploaded to claude.ai isn’t available in Claude Code or the API.

DESIGN.md and design tokens

DESIGN.md, from Google Labs, is “a format specification for describing a visual identity to coding agents”: design tokens (named colour, type and spacing values) in YAML, with Markdown on how to apply them. Its tool flags broken token references and low-contrast colours, and exports the tokens for Tailwind or the format below. Keep it next to the code it styles.

The Design Tokens Format Module 2025.10, marked stable by a W3C community group though not a W3C Standard, is a JSON format that design tools and build pipelines import, so nobody retypes a hex code.

Combining formats

In a 2024 study, GPT-3.5-turbo’s score on one code task varied by up to 40% between plain text, Markdown, JSON and YAML, while GPT-4 was more robust. It wasn’t about brands and its models have since been replaced, so test your own setup. What each format carries points to using several; inside an Agent Skill, all of it can sit in one folder:

One fact can live in two files. The colour reference says when the colour is allowed and why:

## Accent

#D94F30. Use it for one highlight per layout, such as a
button or a single word, and never as a background, so
it keeps drawing the eye.

The token file holds the same value for design tools and code:

{
  "color": {
    "$type": "color",
    "accent": {
      "$value": {
        "colorSpace": "srgb",
        "components": [0.851, 0.31, 0.188],
        "hex": "#d94f30"
      },
      "$description": "One highlight per layout; never a background."
    }
  }
}

Change the two together, or generate the tokens from a single source, as DESIGN.md’s export does. How to version brand guidelines for AI workflows covers keeping the versions straight.

Keep one source current

Every format above stays as it was once copied. The Model Context Protocol (MCP) lets agents read one current source instead: a guide offered as a resource reaches the model when the app includes it, while a tool is called mid-task. Apps can hold on to what they fetched, so have agents fetch at the start of each task.

Rainbrand keeps a brand in the combined form above and serves it to connected agents:

1. Add a brand from PDFs, decks, documents, Markdown, your website, logos, fonts and images. Rainbrand keeps it as a Brand Skill of guidance, values and files.
2. Agents connected through MCP, such as Claude, ChatGPT, Codex and Cursor, request a Brief, the guidance and files for their task, and a Review of the result.
3. When the brand changes, describe the edit and apply the update Rainbrand prepares; later Briefs use it.

Products use the same brand through the API, compared in MCP vs REST API for brand context. An exported folder is a snapshot, so prefer the connection wherever a tool supports it.

Test what your agent actually receives

Before converting anything, give your agent the brand material as it gets it in real work, and ask:

Using only the brand material available to you, quote our rule for using the primary colour and give its exact value. Then name the logo file you would place on a dark background, with its file format, and the font file you would use for headlines. If you can only see a picture of the logo, or are estimating a colour from an image, say so.

Anything it can’t find, sees only as a picture or quotes from an older version needs another format. To stop maintaining copies, sign up and connect your agents to Rainbrand.