Without a brand brief, every AI output sounds like the internet average. With one, AI sounds like you. A brand brief is the highest-leverage document in your AI stack — more useful than any prompt template, more durable than any fine-tune.
Most teams skip it. They write long system prompts, iterate on outputs one by one, and wonder why the voice still feels off. The brief is the fix.
What a brand brief contains
A working brand brief has five sections. Keep each one short. Brevity forces precision.
1. Voice in three adjectives
Pick three adjectives that describe how you write at your best. Not aspirational adjectives — descriptive ones. "Direct, dry, specific" is useful. "Innovative, customer-centric, passionate" is not.
2. Audience in one sentence
Who reads this, and what do they already know? One sentence. "Performance-obsessed growth leaders who can read a diff and are allergic to hype" is a sentence the model can use. "Marketing professionals" is not.
3. Banned phrases
The specific words and patterns you never use. Not categories — exact phrases. "Seamless", "game-changing", "in today's fast-paced world", "unlock your potential". The more specific, the more useful.
4. Content standards
What every piece of content must include. A specific number. A named mechanism. A concrete example. A CTA that does not start with "Discover". Write these as requirements, not suggestions.
5. Three examples of your best writing
Paste them in full. Not excerpts — full pieces. The model needs enough text to pattern-match your rhythm, your sentence length, your paragraph structure.
The negative constraints section is the most important
Positive instructions tell the model what to do. Negative constraints cut off its default paths.
The model's defaults are trained on the internet average. Left to its own devices, it opens with a question, ends with "Discover more", and uses "seamless" at least once. Positive instructions compete with those defaults. Negative constraints eliminate them.
Do not use the word seamless. Do not open with a question. Do not end with a call to action that starts with Discover.
Those three constraints do more work than a paragraph of positive guidance. Write your banned phrases list by reading AI output you hate and extracting the specific words and patterns that made you cringe. That list is your negative constraints section.
Examples beat instructions
Include three examples of your best writing in the brief. The model pattern-matches to examples faster than it follows abstract instructions.
Abstract instruction: "Write in a direct, confident tone with short sentences." Example: a 400-word post you wrote that does exactly that. The example wins every time.
The model can infer sentence length, paragraph rhythm, heading style, and vocabulary from a real piece of writing. It cannot reliably infer those things from a description of them.
Choose examples that represent your voice at its best: the post you are most proud of, the email that got the best response, the proposal that won the deal. If you are unsure which pieces to include, ask yourself which ones you would show a new hire on day one.
How to write the brief in 45 minutes
This is a timed exercise. Set a timer for each section and stop when it goes off.
- 10 minutes: Write your voice in three adjectives and your audience in one sentence. Do not overthink it. You can revise later.
- 15 minutes: List every phrase you hate seeing in AI output. Read a few AI-generated pieces in your category and note what makes you wince. Those are your banned phrases.
- 10 minutes: Write your content standards. What must every piece include? A number, a mechanism, a named example, a specific CTA format. Write them as requirements.
- 10 minutes: Pick three examples of your best writing and paste them in full.
Total: 45 minutes. The brief will be rough. That is fine. A rough brief outperforms no brief by a wide margin.
How to test the brief
Take something you have already written — a post, an email, a proposal section. Ask the model to rewrite it using the brand brief as its only instruction.
Compare the output to the original. Read both aloud.
If you cannot tell which is which, the brief is working. If the output is still generic — still opening with a question, still using "seamless", still ending with "Discover" — add more examples and more negative constraints.
The test is binary: does the output sound like you, or does it sound like the internet? Iterate until the answer is the former.
The brand brief as a business asset
A working brand brief is a business asset. It encodes your voice in a form that any AI model can use, without you being in the loop for every output.
It onboards new tools without you retraining them. When you switch models, add a new agent, or bring on a contractor, the brief travels with the work.
It maintains consistency across a high-volume content operation. At Avakata, 160+ agents produce content daily. The brand brief is what keeps that output coherent.
Treat it like one. Version it. Update it when your voice evolves. Review it quarterly.
We send our Avakata brand brief template — the exact document we use to keep 160+ agents on-brand — to Field Notes subscribers. Get it at avakata.agency/contact.html.
If you want to see how we apply this inside a live agentic content operation, book a discovery call. We will walk through the brief, the agent stack, and what it takes to get consistent output at volume.