You ask an AI tool to draft a launch email or scaffold a landing page. It produces something in the wrong font, using an approximation of your brand blue, and with copy that could belong to any company in your category. So you paste your brand rules into the prompt and ask it to try again. And you repeat those steps next time, and the time after that.
Machine-readable brand guidelines automate that step by providing brand standards rewritten in a structured, unambiguous format an AI tool can read and apply directly, without a person translating first. They're emerging fast because AI is now producing a large and growing share of content, but PDF brand books weren’t designed to give AI tools structured, consistently retrievable brand rules, and support for interpreting their text, visuals, and structure varies between tools.
But a file the machine can read is the beginning, not the finish line. The basics, like colors, fonts, and logo rules, don't carry the real differentiators of your brand, and they don’t solve the system around it: how it gets distributed to AI tools, kept current, and governed as more teams and tools depend on it.
What are machine-readable brand guidelines?
Machine-readable brand guidelines are your brand standards rewritten in a structured, unambiguous format an AI tool can read and apply directly. Instead of a 40-page PDF built for a person to interpret, you get short, explicit rules a model can execute, such as exact hex values, character limits for a headline, a list of words you never use.
Traditional brand guidelines are written for humans. They explain the why behind your brand, with your brand story and long examples to help readers understand your brand. Machine guidelines skip the narrative and skip straight to the what: do this, not that, in a format that can only be interpreted in one way. Each version is optimized differently.
Machine-readable brand guidelines aren’t replacing your human ones. They’re a second type of documentation. One version helps people understand the thinking behind your brand, and the other tells tools exactly how to execute it.
This is the brand equivalent of a movement already reshaping how AI tools work. Just like AGENTS.md gives coding agents instructions for a codebase, machine-readable brand guidelines give AI tools the context to stay on-brand without you repeatedly providing that information.
Why brand guidelines have to be machine-readable now
According to Salesforce, 75% of marketers have adopted AI. And it’s not just marketing teams using it. Gen-AI tools help teams across the business, from sales to HR, create content that gets shared outside your organization.
Imagine ten teams across four regions, all using different AI tools to generate brand content.
Each team has access to your brand guidelines, but those guidelines aren’t necessarily connected to the tools they’re using. The result is that generated content doesn’t consistently follow your brand, and any ambiguity in the guidelines can lead to different interpretations.
To avoid these inconsistencies, team members can manually explain your brand palette, font, and voice in every new AI chat. But that repetition creates a recurring cost in time and work across every person and AI interaction. A machine-readable set of guidelines that can be used as a repeated source of truth removes much of that manual work.
Machine-readable guideline formats
There are several different formats for machine-readable brand guidelines. These aren’t necessarily competing approaches, but rather different tools for different jobs, so a mature brand may use more than one.
brand.md — an emerging open specification
brand.md is an emerging, MIT-licensed open specification created by Caio Pizzol for writing your brand down in a structured format AI tools can use as context. It's a single text file, written in Markdown, that lives in your AI project's "root directory," alongside similar files like README.md.
The brand.md file is organized into three layers:
At the very top of the file sits something called YAML frontmatter, a short block of basic details (your brand name, tagline, version number, brand type and brand architecture) written in a simple format that both humans and machines can understand instantly.
For global and enterprise companies, brand.md supports brand hierarchies, so you can have master brands with sub-brands that inherit (or sometimes override) various brand elements. It also supports different brand architectures, which makes it useful for multi-brand organizations.
Because it's just plain text, brand.md needs no special software to run. It displays cleanly on GitHub, looks readable in any code editor, and can be used as context by AI tools that are configured to load or receive it. It's a great starting point for creating brand guidelines in a format AI tools can work with.
Brand skills (SKILL.md)
A "skill" is a package of instructions built around a SKILL.md file that compatible agents can discover and load when it's relevant to the task at hand. The format was originally developed by Anthropic and is now an open standard. That's the core difference from brand.md: a brand.md file is a static document available as context, while a skill is more like a capability an agent reaches for on demand.
As an example, you'd package your color palette, typography, voice, and instructions for when to apply each element into a skill called something like brand-guidelines. Then, whenever you ask a compatible agent to build a presentation, write a report, or format a spreadsheet, it checks whether a relevant skill exists, loads it, and applies your brand without having to re-explain your rules in the prompt.
There are some specific mechanics that differentiate it from brand.md:
That split between instructions and scripts supports brand governance. Anything genuinely non-negotiable, such as an exact hex code, a logo clear-space ratio, or a banned word, can be encoded in a script, which reduces interpretive drift by applying exact values or logic. Anything that requires judgment, such as how formal to sound in a given context or which of three approved photography styles fits a mood, belongs in an instruction, where the model has room to apply nuance. The surrounding system still needs to select and run the right script for those rules to be enforced.
While a single brand file is a good starting point, skills can be more powerful, especially for enterprise organizations with lots of brand-related documentation:
- They're composable: Skills can reference each other. Your brand-guidelines skill can work alongside a product-research skill or a writing-standards skill, so a single request can pull in brand formatting and accurate product facts and your house writing style, all at once.
- They travel with the work, not the prompt: Once installed, a skill doesn't need to be pasted in or reattached. It's simply there, ready to use when needed.
design-md (DESIGN.md) — for design and code tools
brand.md and skills are built for content — copy, decks, documents. design.md (sometimes called design-md) is built for tools that generate interfaces and code.
It's an open-source draft specification that Google Labs released through Stitch, its AI design tool. Like brand.md, it's a text file, but instead of positioning statements and tone-of-voice rules, it pairs "design tokens", exact, reusable values like a specific color code or font size written in YAML, with a short plain-text note explaining why that value exists. A tool reading the file gets both the number and the reasoning behind it.
Design-and-code tools can use a design.md file when it's provided or integrated into their workflow to generate interfaces and components that match your brand. Stitch supports the format directly, and Google also has a companion command-line tool that can check a design.md for accessibility issues.
Both brand.md and design.md are still emerging conventions. DESIGN.md is currently at alpha, so it’s one to watch as it matures, rather than something to build an entire system around today.
How this relates to AGENTS.md
AGENTS.md follows a similar principle to machine-readable brand guidelines: it gives AI tools structured, plain-text instructions they can use directly, rather than relying on someone to explain the same context in every prompt. The difference is what that context is for.
- AGENTS.md lives in a codebase and gives coding agents, such as Claude Code or Cursor, instructions for working within that project. This might include how to run tests, which conventions to follow, or what not to change.
- brand.md, brand skills, and DESIGN.md provide context about your brand instead, including how it looks, sounds, and should be applied.
What goes into machine-readable brand guidelines
Most machine-readable brand files only provide information on colors and fonts. The output looks on-brand, but the content still feels generic because the AI tools have no idea what to say. The best machine-readable guidelines go much deeper, providing more context about your brand:
From a single file to a brand system: the architecture decision
When you’re creating machine-readable guidelines, there are different levels of documentation you could create. How far you need to go depends on your company’s size, number of brands and markets, and how many people and AI tools touch the brand. Here’s an overview of the three levels of guideline documentation, starting with the simplest.
A single brand.md file
For a small team with one brand and a handful of people generating content, one file is genuinely enough. It's low overhead — write it once, drop it in your project root, done — and it starts working immediately, with no infrastructure to stand up.
But as companies grow, they find a single file has three clear limits:
- Distribution: A file that lives in one project root doesn't automatically reach every tool, every team member, or every AI agent your company uses. Someone has to know it exists and manually attach it each time.
- Version drift: The moment you update your voice guidelines or add a new product color, every copy of that file elsewhere is instantly outdated.
- Depth: A single file has a practical ceiling. Try to cram in multiple product lines, regional variants, and detailed application guidance, and it becomes unwieldy.
None of this makes a single brand.md a bad choice, but many companies outgrow this single file approach relatively quickly.
A structured set
The natural next step is more of the same, with a clear structure to the information. Brand.md becomes the overview file, with more detailed files referenced beneath it: one for voice, visual identity, product line, or region. Alternatively, if you're working with a skill instead of a static file, you get the same effect through progressive disclosure — a short top-level description, with fuller detail and reference files pulled in only when a task actually needs them.
This means larger brands with several product lines or markets can now represent the complexity of their brand without cramming it into one file. In workflows that support selective retrieval or progressive disclosure, AI tools can load only what's relevant to the task, keeping that context efficient instead of bloating every prompt with sections that don't apply.
However, while it lets you go into more detail than a single brand.md file, a structured set of files inherits, and in some ways compounds, the distribution and version drift problems.
Served, governed brand context
Served, governed brand context means your brand guidelines don't live as a file that has to be found, copied, and attached. Instead, they live in one central system and can be retrieved by whichever tool or person needs them, in the right form, at the right moment.
The delivery mechanism is typically something like an API or an MCP server: a way for AI tools to connect to and retrieve current data from a central system, rather than reading a static file someone manually copied in. This offers several benefits:
If you’re using a single brand.md file or a set of structured skills, those still need to be synced and updated to align with your internal documentation as your company brand evolves. Served, governed brand context can reduce that manual work when it pulls brand information from your central source rather than relying on copied files. This structure can address the challenges of distribution, updates, maintenance, role-based navigation, and governance without relying on your team to manage every copy themselves.
The importance of task- and role-based navigation in machine-readable guidelines
The same brand system has to support the execution of a wide variety of tasks:
- "Write a LinkedIn post for this product and audience" is a narrow, single-output task
- "We have a March launch, help me plan the campaign" is a multi-step planning task
- "Automate our whole content production with review workflows" is a standing system, not a one-off request.
All three requests would use the same underlying brand guidance, but they’d need different amounts of detail and depth.
But there’s a second variable that introduces more complexity to your brand guidelines. People in different roles also need different information from your system:
- A content marketer drafting social copy needs voice rules and message pillars, not the full visual identity system
- A designer needs tokens and imagery rules, not prose about tone
- A PM briefing an agency needs an overview of brand rules
- A CMO auditing quarterly output needs something else again.
Your brand guidelines are accurate and complete, but they never reach the person in the right moment. Machine-readability doesn't fix that on its own, but the right routing makes those guidelines more usable and relevant across your business.
A well-built system can retrieve correct information most of the time. But routing is more complicated — when a team member is using an AI tool to generate brand content, the AI tool knows the task and the person’s role, and it needs to work out what parts of your brand guidelines are relevant and applicable to that use case.
Getting that right, automatically, for every combination of task and role can’t be done with a simple brand.md file. Those files deliver the same brand information to every user, no matter who they are or what they’re trying to do. To achieve this level of segmentation, you need a system that can identify the task, identify the role, and provide only the relevant slice of guidance for that combination.
When and how to keep humans in the loop: orchestration, approval, and observability
The aim isn’t to replace your human brand team with a set of machine-readable guidelines and have AI manage your brand. Instead of trying to remove people from the process, you want to train your AI so it gets better at knowing when to route work to a person, and exactly who’s the right person for different queries.
Keeping humans in the loop, even with the most detailed machine-readable guidelines, is essential because:
- AI models shouldn’t make important judgement calls. Legal risk, sensitive claims, and brand-defining decisions need a person who owns that judgment and is accountable for it.
- AI models earn trust over time. You shouldn’t hand over full content production to an unsupervised AI system on day one. Your team needs to review outputs and extend autonomy over time, as the system demonstrates it can consistently produce on-brand materials without significant corrections.
- Your team needs to maintain accountability. If you publish off-brand or that creates legal risk, you can’t just say “the AI did it”. Keeping your team in the review and sign-off process means there’s always a real person who has ownership of those final decisions.
- Today, many companies have a governance gap when using AI tools to create brand content. AI agents often run on someone’s own machine, which isn’t scalable because it’s not properly governed by your brand or organization.
Today, many companies have a governance gap when using AI tools to create brand content. AI agents often run on someone’s own machine, which isn’t scalable because it’s not properly governed by your brand or organization.
What this looks like in practice
To solve this challenge, you need to properly orchestrate the connection between brand information, AI tools, and the people in your company:
- Index people and roles, not just brand guidance. Your machine readable guidelines will cover your brand rules as standard. Give AI tools more context by defining the people and stakeholders in your organization who own important decisions, like who signs off on legal claims, or who approves brand-critical campaigns. Then your AI tools can suggest reaching out to the right person at the right step, instead of generating content and pushing it forward unchecked.
- Build in auditable checkpoints. The system should recognize which steps call for human judgment and pause there, rather than treating every task as fully automatable end to end. It should also keep a record of who signed off on any content, and when, so you can keep an audit trail that users can check later.
- Route AI-generated artifacts back into a shared system. Instead of running AI agents against your own brand guidelines on people’s local machines, you need a shared system that captures what was generated, what was approved, and what got flagged, so the work is visible across the organization.
Companies often assume their AI adoption is a success when the system can produce on-brand content. But we believe a more important success measure is whether, when the AI hits a decision that it doesn’t have the context or authority to make, it knows to stop and ask the right people — and whether anyone else in the organization can see that it did.
How Frontify makes your brand machine-readable — served, governed, and connected
While a single brand.md file is the common starting point for companies creating machine-readable brand guidelines, mature, global companies need a system designed for the complexity of their brand and organization.
The Frontify platform hosts guidelines, assets, and brand knowledge in one central location, structured so it’s easy for the people in your organization to navigate and use — and also readable by AI. Guidelines in Frontify can be served to whatever AI tool or agent needs it, rather than rely on your team to copy and paste it in. Here’s an overview of how the Frontify platform makes your brand machine-readable, while maintaining proper governance.
Connected via MCP. Model Context Protocol (MCP) is a standard way for AI systems to connect to external tools and data sources. Frontify’s MCP server connects brand guidelines, assets, and templates directly to AI tools like Claude, ChatGPT, Cursor, and Figma, so a request in any of those tools can pull real brand context from your central platform.
Delivers role-specific brand context. Frontify’s MCP server runs through each user's existing Frontify permissions, so a team member using AI to generate brand content only sees the brand assets and guidelines that relate to their role, region, or permissions level.
AI-discoverable visual assets and text. Frontify auto-tags assets on upload using image recognition, applies predictive metadata, and can extract existing metadata from a file’s embedded data. That turns images and visual content into something AI tools can search, filter, and retrieve — just like text.
Answers brand questions with sources, routed by permission. Frontify’s Brand Assistant is a conversational layer over your brand guidelines. Users can ask questions and it provides natural-language brand answers sourced from the guidelines, with source links. The MCP and API connections mean Frontify’s Brand Assistant is also reachable by the other AI tools in the workflow, not just inside the Frontify platform itself. It also respects each user's access permissions, so its answers vary depending on who's asking.
Visible brand governance. Instead of agents running locally and invisibly, Frontify’s permissioned access, approval workflows, and usage visibility mean brand and legal leads can see how AI is actually touching brand assets — which agent did what, when, and whether it was inside the access it was granted.
See how Frontify brings brand guidelines into your existing AI workflows: book a demo with our team.
