How I stopped copy-pasting brand context into every AI prompt

This started as a workflow problem.

I was doing brand work with AI tools — ChatGPT and Claude mostly — and every new conversation started the same way: copy the brand colors, paste the font names, write out the tone of voice rules, add what words to avoid. Sometimes 400 words just to set context. And maybe 60% of outputs actually felt on-brand. The AI would get the colors right but miss the tone. Or nail the tone and get the terminology wrong.

I kept thinking the AI wasn't good enough. Then I started looking at how I was feeding it the information.

The format problem

AI doesn't process brand guidelines the way a human does. It processes structured information. When you paste a wall of brand guideline text into a prompt, you're asking the model to extract structure from something unstructured — it guesses. Inconsistent input, inconsistent output.

Longer prompts mostly don't fix this. Better structure does.

What I did instead

I started restructuring the same brand info as a markdown file — hierarchical, with explicit sections:

# Brand Identity
Company: YourCo

## Color Palette
- Primary: #2C3E50 (Deep Navy) — headlines, CTAs only
- Accent: #E74C3C (Signal Red) — alerts only
- Background: #FAFAFA

## Typography
- Display: Playfair Display Bold
- Body: Inter Regular 16px
- Code: JetBrains Mono

## Voice & Tone
- Direct, not corporate
- Confident without being arrogant
- Active voice always
- Avoid: "leverage", "synergy", "game-changing"

## Messaging
- Lead with the problem, not the product
- Proof before claims
- CTAs: action verbs only ("Try it", "See how it works")

Then I'd load it as a Claude Project reference file, or paste it into a GPT's system instructions. The difference was immediate. Tone consistency improved right away — the AI started avoiding banned phrases without being reminded. Color and typography references became reliable. Fewer correction loops.

How to deploy it

Claude Projects: Add the file as a Project resource. Every conversation in that project inherits the brand context automatically — no copy-paste required.

ChatGPT: Paste the file content in a custom GPT's system instructions, or add it as an attachment to a conversation.

API / system prompts: Include it in your system message. Works for any application you're building on top of these models.

The extraction problem

The hard part isn't using the file — it's building it from existing brand assets. If your guidelines live in a PDF, Figma file, or PowerPoint deck, you're manually translating them. That's tedious, especially when you're handling multiple brands.

Doing this by hand for every client got old fast, so I built brandMD to automate the extraction step. Upload a brand PDF, it parses it and outputs a structured markdown file ready for any AI tool. $9.99 one-time, processed client-side (nothing leaves your browser).

You can explore full extractions for real brands — Anthropic, Slack, and others — so you can see what the output actually looks like before committing.

The manual approach works too. If you're working with one brand and have the time, 30 minutes to build the file once is worth it. The consistency improvement is real.

Happy to answer questions or share more specific examples of what the structured files look like for different brand types.