How to Brief an AI on Your Brand (the Right Way)

To brief an AI on your brand, give it structured brand files at the start of every session — not a description of your brand, and not a PDF of your guidelines. The brief is a set of documents the model can parse: purpose, positioning, audience, voice rules, owned vocabulary, banned language, and the boundaries of what the brand will never say.

The distinction sounds pedantic. It is the entire difference between output you rewrite and output you ship.

Why the PDF doesn't work

The instinct is reasonable: you paid for brand guidelines, so you upload them. Three things go wrong.

The format fights the machine. A designed PDF is built for human eyes — columns, mood boards, images of the logo on tote bags. Text extraction from that layout is lossy and arrives out of order. The model receives fragments.

The content isn't executable. Guidelines typically say things like "our tone is confident yet approachable." That's a description, not an instruction. A model can't act on it, so it falls back on the average of everything it has read about confident-yet-approachable brands — which is to say, everyone.

The brief evaporates. Even when it partly works, you're doing it again tomorrow, and in the next tool, and in the tool after that.

A useful principle:

An AI brief isn't a description of your brand. It's the set of decisions your brand has already made, written so they can be applied without you in the room.

What an AI-readable brief actually contains

Five components, ordered by how much they change the output.

1. Owned vocabulary. The specific terms your brand uses, with definitions. Proprietary language is the single strongest signal you can hand a model, because it cannot be averaged with anyone else's writing.
2. Banned language. The words and phrases you never use, ideally with replacements. Models follow explicit prohibitions remarkably well — better, in practice, than they follow aspirational descriptions.
3. Positioning boundaries. What the brand refuses to claim, sell, or promise. Refusals are executable in a way that values statements are not. "Never promise revenue outcomes — promise outputs" changes drafts immediately. "We value integrity" changes nothing.
4. Audience specifics. Not demographics. The situation your reader is in, the language they use for their own problem, what they've already tried, what they're skeptical of.
5. Structural rules. How a piece opens, how it closes, sentence rhythm, whether questions are allowed, how the call to action is phrased.

Each item should be a short declarative rule. Prose about your values is decoration; rules are instructions.

The mechanics, tool by tool

The delivery method differs slightly by platform, but the pattern is identical: put the files where the model reads them before the first message, not inside it.

• Persistent workspaces (project files, knowledge folders, custom instructions): load the brand files once; every conversation in that space starts with the context already present. This is the best available option in most tools today.
• Single sessions: attach the files as documents at the top of the conversation. Attaching beats pasting — attached files stay legible to the model across a long thread, while pasted text drifts backward and loses weight.
• API and custom co-pilots: the same files become the system context. Nothing new to write.

Keep the master copies in plain Markdown in your own storage. Every tool ingests plain text natively, which means one set of files serves all of them — and follows you when you change tools. That portability is the difference between owning your brand context and renting it.

The correction loop (where the compounding happens)

Briefing is not a one-time act. It's a loop, and the loop is where the asset gets built:

1. Load the files. Generate.
2. Read the draft and notice what's off.
3. Fix the file, not just the draft. Word you'd never use? It goes on the banned list. Structure that felt wrong? Write the structural rule.
4. Next session starts from the improved brief.

Most founders skip step 3 and re-edit the same errors for a year. This is what I'd call Prompt Waste: paying repeatedly, in your own attention, for a fix you could have made permanent in thirty seconds.

What a brief must never do

Never ask the AI to invent the brief. "Create a brand voice for me," "define my positioning" — these requests return the average, formatted persuasively. The model can extract, structure, and challenge what you already know; it cannot originate your purpose, because it has no access to your experience. The direction is always human to machine.

Never let the brief live only inside a platform. If your brand context exists exclusively in one vendor's settings panel, you don't have a brief. You have a dependency.

Where to start

Write the banned-language list first. It takes twenty minutes, it's the easiest thing to be certain about, and it produces a visible change in the very next draft. Add the owned vocabulary next. Everything else can accumulate through the correction loop.

If you want the full structure — all the decisions a brand needs documented before an AI can execute them — that's the Brand DNA work we do at openidea.biz. If you'd rather start by seeing where your own foundation stands, the free Brand Foundation Check runs the diagnostic in about twelve minutes.