Design & Tools
7 min read

Brand guidelines for AI: what n8n's design file gets right and how to write your own

Robot arm made of toy bricks placing a pink brick on a colorful wall, six bricks beside it as a color palette
Image: AI-generated
In short

Brand guidelines for AI are a short text file with exact color, type and spacing values plus behavior rules that Claude, Cursor or the Figma Agent read before any design task. Since its brand refresh in October, n8n offers exactly this kind of file. What matters most are written rules for contrast, components and gaps, because an AI fills in anything that is missing on its own.

What are brand guidelines for AI?

Brand guidelines for AI are a text file that an AI tool reads before every design task: colors, fonts, spacing, logo rules and things to avoid, written as exact values instead of a mood board. The classic brand PDF is written for people who read between the lines. A language model does not read between the lines. It fills gaps with whatever is statistically most likely, and the result looks like every other SaaS website.

If you build landing pages with Claude or Cursor, create social graphics with the Figma Agent or generate pages with Webflow Campaigns, you need a second version of your brand rules: machine readable, short and unambiguous.

What did n8n change about its brand on October 14?

On October 14, n8n launched a refreshed identity at its In The Loop conference. It is a brand refresh, not a rebrand: the brand gets reworked while its core stays the same. Jan Oberhauser calls it "a zhush, not a rebuild" in the blog post from October 20.

Changed Kept
Logo and wordmark: show activity instead of connection The pink, only slightly refined
New typeface Playful character
Added grays so the pink no longer dominates everything Community Edition and free choice of models

The interesting part is not in the blog post but in the new brand guidelines. Next to tone of voice, logo, colors, typography, grid and graphic elements, there is a separate section called "Build with AI". There n8n offers the file n8ndesign.md for download, version 5 from October 19. It is explicitly meant for Claude, ChatGPT, Codex and Cursor. The instructions recommend anchoring the file in the agent's instruction file (for example with @n8ndesign.md in CLAUDE.md) and having the agent read it before any UI work. Sections such as motion and imagery are still marked "Coming soon".

Our take: this is the real shift. Brand guidelines now have two readers, and the second one works faster and more often than any agency.

What belongs in AI-ready brand guidelines?

An AI tool needs two layers: exact values it does not have to interpret, and short explanations of when each value applies. That is exactly how DESIGN.md is built, an open format from Google Labs (Apache-2.0 license). The values sit as design tokens in the YAML header of the file. Design tokens are named design values such as primary: #FF91AC or spacing-m: 16px that every component refers to. The prose below explains the intent.

The specification defines these sections in a fixed order:

  1. Overview: what does the brand stand for, how should it feel?
  2. Colors: primary, accent and neutral colors with hex values and purpose
  3. Typography: fonts, sizes and line heights for each text level
  4. Layout: grid and spacing scale
  5. Elevation & Depth: shadows and layers
  6. Shapes: corner radii and shape language
  7. Components: buttons, cards and form fields with their states
  8. Do's and Don'ts: the mistakes you have already seen

One thing to plan for: the format is at alpha, and its makers say themselves that it will still change. If you create a DESIGN.md today, treat it like code, with a version number and a date. n8n shows how.

Where do AI brand rules fail in practice?

They fail because of what is not in them. Three typical gaps:

Contrast is missing as a rule. By our calculation, the n8n pink #FF91AC on the light background #FAFAFA reaches a contrast ratio of about 2.0:1. WCAG requires at least 4.5:1 for normal text and 3:1 for large text (18 point and up, or 14 point bold). On the dark #191919, the same pink reaches about 8.3:1. On the colors page of the n8n guidelines we found no explicit contrast rule on October 22. A person on the design team knows that pink on white is not for body text. An agent that reads "pink as an accent" may well put a pink link on a white background. So write these rules out: "Never use the accent color as text color on a light background."

Gaps get filled creatively. n8n solves this with a line you should copy: "If something isn't covered, ask me instead of guessing." Plus: "Reuse its colors and type styles; don't invent new ones." Without sentences like these, the AI invents in-between shades, and three months later you have twelve shades of blue.

Prose instead of values. "Generous white space" is not an instruction for a model. "Section spacing 96 px on desktop, 64 px on mobile" is.

How do you write your own brand guidelines for AI?

You do not need a new corporate design, just a translation of the one you have. This is how we approach it:

  1. Collect the values: pull hex codes, font names, font sizes, spacing and radii from Figma variables or your website's CSS. Do not copy them from the PDF, which is often outdated.
  2. Note one purpose per color: background, text, accent, error, success. Add every allowed text and background combination with its contrast value.
  3. Describe components with their states: button default, hover, disabled. This is where most outliers come from.
  4. Write do's and don'ts as scenarios: situation, right, wrong, reason. Figma recommends the same for its agent guidelines.
  5. Add behavior rules for the AI: no new colors, ask when something is missing, pick the logo variant based on the background.
  6. Have it checked: the DESIGN.md command line tool has a lint command that, according to the project page, checks broken token references and WCAG AA contrast, among other things. export turns the file into a Tailwind config or tokens in the W3C format.
  7. Test it: give the AI three typical tasks (hero section, social graphic, form) and hold the results against the rules. Every mistake becomes a new line in the don'ts.

Where exactly do you use the file?

A well maintained file works in several tools, but each one has its own rules:

Tool How to add it What you need to know
Claude Code, Cursor, Codex Reference in the project's instruction file This is how n8n describes it for its own file
Figma Agent Guidelines files on a published library Open beta; Professional, Organization, Enterprise; .md, .txt, .json; recommended 3 to 8 files under 100 KB, maximum 200 KB per library
Webflow Campaigns Upload a design.md Separate product with its own price, details in our article on Webflow Campaigns

Two details on Figma from its help page: the agent receives the guidelines automatically as soon as someone designs with the library. They are currently not used for Make Kits. And according to Figma, duplicate content across several files costs more processing time and more AI credits. Since October 6, the agent bills through AI credits, and our article on Figma Agent credits shows how far they go. Short guidelines without repetition are therefore also a cost question.

What does this mean for your company?

If someone on your team already builds slides, graphics or pages with AI, brand appearances are being created right now that nobody approved. AI brand guidelines are the cheapest lever against that: written down cleanly once, they work in every tool that reads them.

  • No brand refresh needed: the file translates your existing design. A refresh like n8n's is only the right occasion if you are reworking the brand anyway.
  • One source, not three: maintain the values in one place (Figma variables or DESIGN.md) and derive the rest. Otherwise the PDF, the website and the AI file drift apart.
  • Assign ownership: someone has to update the file with every design change, with a date and a version.
  • Results still need review: the file lowers the error rate, but it does not replace a look before publishing. For AI images, the labeling requirement applies as well.

Our take: for most mid-sized companies this is translation work on existing values, not a relaunch. If you would rather not translate your brand for AI yourself, get in touch: we can help with design and design systems. If you want your slides on brand from the start, our presentation design services cover that too.

Frequently asked questions

Does a DESIGN.md replace the classic brand book?

No. The brand book explains attitude, imagery and exceptions to people. The AI file is a condensed, machine readable version of the fixed values and rules. Both should be maintained from the same source.

How long should brand guidelines for AI be?

Short enough that every line settles a decision. For its agent guidelines, Figma recommends three to eight focused files with a combined size under 100 KB. Repetition there costs processing time and credits.

Does this work with image generators too?

Only to a limited extent. Hex values and spacing mostly help tools that generate code or layouts. For image generators you also need described imagery templates, and n8n itself still marks this section as Coming soon.

Can I use n8ndesign.md as a template for my brand?

As a structural model, yes, but the content belongs to the n8n brand. A cleaner start is the open DESIGN.md specification from Google Labs, filled in with your own values.

Sources

  1. n8n: Why we refreshed the n8n brand
  2. n8n Brand Guidelines: Build with AI
  3. Google Labs: DESIGN.md specification (GitHub)
  4. Figma Help: Add guidelines files to a design library
  5. W3C: Understanding WCAG 2.2 Contrast (Minimum)
This article was created with AI.
Got a project in mind?

Tell us what you're planning. You'll get an honest assessment within 24 hours.

Got it! We will get back to you asap!
Whoopsie! Something went totally wrong here... Maybe just send an email to hi@blocksandcolors.de? Sorry!
Or email me directly: hi@blocksandcolors.de