For five years I looked after a brand used by more than a hundred offices. We had a good brand book. Sixty pages, beautifully set, signed off by the board. I can count on one hand the people outside my team who opened it.
That wasn’t their fault. The PDF was in a drive, three folders deep, and it answered questions nobody was asking. Someone making a flyer at 4pm on a Thursday doesn’t want a philosophy of the brand. They want to know which green, which photo, and whether they can say “urgent”.
Who reads your brand book now
Here is what changed. Most first drafts are now written by a model. Captions, emails, ad variations, event pages, the alt text on a photo. The person still edits and approves, but the words and often the layout started somewhere that has never seen your PDF.
So the most frequent reader of your guidelines isn’t the junior designer any more. It’s ChatGPT, Claude, Cursor, Canva’s writing tools and whatever your agency built last month. And none of them can open a PDF in a drive they don’t have access to.
A brand book nobody can read is just a very expensive opinion.
What replaces it
Not a better PDF. A brand system: one source of truth that people can browse and tools can query. In practice that means four things.
- It lives at an address. Your brand book is a page with a URL, not a file. When the green changes, it changes everywhere, and nobody is working from version 3_final_FINAL.
- It’s structured. Colours are hex values with roles, not a swatch in an image. Voice is a list of words to use and words to avoid, not a paragraph about being “warm yet bold”.
- Tools can ask it things. An agent can fetch your rules before it writes, and check what it wrote before it shows you.
- It checks the work. Every draft, upload and generated asset gets a score and a list of what to fix, against your actual rules.
Start from what you already have
The objection I hear most is time. Nobody has six weeks to rewrite the guidelines. You don’t need to. Your website already is a brand book, just an unwritten one. It has your logo, your colours as the code actually uses them, your fonts, your headlines and your photography.
That’s where Brandbook starts. Paste your URL and it reads up to 24 public pages, then pulls out the logo, palette, type, voice and imagery into a system you can edit. If you have an old PDF, add it as a source; where the PDF and the site disagree, it asks you which one is right instead of guessing.
You’ll still want to fix things. The scan finds what you do, not always what you meant. But editing a draft is a morning’s work. Starting from a blank page is a quarter.
One key, every tool
Once the system exists, connecting it is the easy part. Claude Code, Claude Desktop, Cursor and VS Code talk to Brandbook’s MCP server with one URL and one key, so an agent can pull your voice rules and run a check without you pasting anything. Your own code can use the REST API. For everything else, there’s a brand.md file you can drop into a ChatGPT or Claude project, and a tokens.json for Figma and your front end.
What stays human
None of this replaces taste. A rule engine can tell you a headline has three exclamation marks and an off-palette red. It can’t tell you the idea is boring. That’s still your job, and honestly it’s the good part of the job. The point of a brand system is to stop spending your week on the green and start spending it on the idea.
The PDF had a good run. It was the right format when brands were made by a handful of people in one room. They aren’t any more. Write your brand down somewhere the machines can read it, and keep the people in charge of what it says.
