CLAUDE.md for designers: the file that keeps AI on-brand

Published:
November 15, 2026
Updated:
November 19, 2026

Claude Code forgets your brand every time you start a session. Most designers fix it by repeating hex codes in every prompt. A 30-line CLAUDE.md does it once, if you write rules instead of brand philosophy.

CLAUDE.md for designers

Every Claude Code session starts from zero. It doesn't remember your brand colors, your button rules or the fact that you hate purple gradients.

CLAUDE.md fixes that. It's the file Claude reads at the start of every session, before your first prompt. Most designers either skip it or fill it with a brand manifesto the model ignores. A short, specific one will keep your output on-brand better than anything you type into the chat.

What CLAUDE.md is

A Markdown file of instructions that Claude Code loads into every session. You write it once, and Claude starts every conversation already knowing your project.

It lives in one of three places:

  • ./CLAUDE.md in your project folder. Shared with your team through Git. This is where brand rules go.
  • ~/.claude/CLAUDE.md in your home folder. Your personal preferences across every project.
  • ./CLAUDE.local.md in your project, kept out of Git. Your own notes for this project only.

All of them load together. If you start Claude Code inside a subfolder, it also picks up the CLAUDE.md files in the folders above it.

Claude Code documentation page about CLAUDE.md and memory
Source: Claude Code docs

Start with /init, then cut

Run /init in a new project and Claude writes a starter CLAUDE.md from what it finds in your files. It's a decent first draft for developers. For design work, it misses everything that matters: your visual rules, your voice and what "on-brand" means to you.

Keep the build commands it finds. Write the rest yourself.

Import your DESIGN.md in one line

If you have a DESIGN.md, don't paste it into CLAUDE.md. Import it:

@DESIGN.md

That one line pulls the whole file into every session. A few details from the docs worth knowing:

  • Paths are relative to the CLAUDE.md file, not to where you started Claude.
  • Imported files can import other files, up to four levels deep.
  • A path with spaces needs a backslash before each space, like @Brand\ Docs/voice.md. Quotes don't work.
  • Importing a file from outside the project triggers a one-time approval prompt.

I keep my own voice and content rules in separate files and pull them into any session that touches copy. One source, every session.

Write hard rules, skip the philosophy

Anthropic's own guidance: the more specific and concise your instructions, the more consistently Claude follows them. "Our brand is bold yet approachable" gives the model nothing to act on. "Headings in sentence case" does.

Vague, ignored Specific, followed
Keep it clean and modern Max 2 font weights per page: 400 and 700
Make CTAs stand out One primary button per screen. Secondary actions are text links.
Write in our voice No em dashes. Contractions always. Headings in sentence case.
Be accessible Body text at 4.5:1 contrast minimum. Every image gets alt text.

A CLAUDE.md you can copy

A starting point for a marketing site project. Swap in your own rules.

# Project: Acme marketing site

## Design system
@DESIGN.md

## Voice
@brand/voice.md

## Hard rules
- One primary CTA per screen. Everything else is a text link.
- Headings in sentence case. No title case anywhere.
- Never invent stats. Every number needs a linked source.
- External links open in a new tab.
- Images: WebP, under 200KB, always with alt text.

## Before you finish
- Check heading order: one H1, no skipped levels.
- Check every page has a meta title and description.

Under 30 lines. That's the point.

Keep it under 200 lines

The docs recommend staying under 200 lines per CLAUDE.md. Longer files eat context and Claude follows them less closely. Imports don't help here, because imported files load with the file that imports them.

If some rules only apply to certain files, move them to .claude/rules/ and scope them with a paths field. A rule about HTML templates then loads only when Claude touches HTML.

Watch for contradictions

If two instructions disagree, the docs say Claude may pick one arbitrarily. Your DESIGN.md says buttons are fully rounded, an old line in CLAUDE.md says 8px radius, and the output flips between the two.

Every time you change a brand rule, search all your instruction files for the old version. Or let Claude do it: run /doctor prompt-audit and it flags contradictions, stale rules and references to files that no longer exist. It proposes the edits and changes nothing until you say so.

Read what Claude remembers about you

CLAUDE.md is what you tell Claude. Auto memory is what Claude writes down on its own: notes from your corrections and preferences, kept per project. The first 200 lines load into every session, right next to your rules.

Most of the time that helps. But if you corrected it once on a one-off project ("make this one purple"), it can learn the wrong lesson. Open /memory now and then, read the notes and delete anything that isn't true anymore.

When a rule must never break

CLAUDE.md is guidance. Claude treats it as context, and it can still slip. For something that should never happen, like publishing to production without your approval, use a hook or a permission setting. Those block the action outright.

For everything else, a tight CLAUDE.md does the job. Write it once, fix it when the output drifts, and stop repeating your hex codes in every prompt.

Book a free design audit → humbldesign.io

Any statistics cited in this post come from third‑party studies and industry reports conducted under their own methodologies. They are intended to be directional, not guarantees of performance. Real outcomes will depend on your specific market and execution.

Follow David on LinkedIn
More on AI, Webflow, and Claude + MCP workflows.
David Pokorny
1

What is CLAUDE.md?

CLAUDE.md is a Markdown file of instructions that Claude Code loads at the start of every session. You write your project rules once and Claude starts every conversation already knowing them.

2

Where should I put my CLAUDE.md file?

Put team and brand rules in ./CLAUDE.md in your project folder so they're shared through Git. Personal preferences for every project go in ~/.claude/CLAUDE.md, and private notes for one project go in CLAUDE.local.md.

3

How do I add my DESIGN.md to Claude Code?

Add a line with @DESIGN.md to your CLAUDE.md. Claude Code expands the import and loads the full file into every session. Paths are relative to the CLAUDE.md file.

4

How long should a CLAUDE.md be?

Under 200 lines, per Anthropic's docs. Longer files use more context and Claude follows them less closely. Move rules that only apply to some files into .claude/rules/ with a paths field.

5

Why is Claude ignoring my CLAUDE.md rules?

Usually because the rules are vague or contradict each other. Rewrite them as specific, checkable instructions and search your instruction files for conflicting versions. For rules that must never break, use a hook or permission setting instead.