The First Button of Any AI Marketing Agent: A Brand Brief, Not a Smarter Model

Audience: marketers · solo operators · agent builders  ·  Format: workflow + methodology  ·  2026-06-24

Image: From one brief, multiple agents fan out in the same direction.

"Build me a marketing strategy" produces output that fits no brand

Start with one scene. You tell a marketing agent "build an ad strategy for this company," and polished output arrives in five seconds.

It outlines a funnel, sets a persona, sketches a channel mix, all plausible. But when you read it line by line something is off. The company is a 30-year-old offline wholesale distributor, yet the output is talking about app push notifications and retargeting pixels.

The model isn't stupid. It had no information, so it filled in the most common defaults. AI doesn't leave blanks empty. It fills them with the most plausible average.

The result fits any brand, and for exactly that reason it fits none. This isn't solved by upgrading the model. It is an input problem.

Under the name of Bigyowon (a Korean telecom and rental comparison service), I am running a marketing agent team where one person covers market research, positioning, growth, content, and data.

The most expensive mistake I've learned from running that team is not "a wrong answer." It is "a confident answer built on guessed input."

A wrong answer is visible and gets discarded. But a strategy precisely constructed on top of plausible average values only reveals "this wasn't about us" after a long stretch of execution. That is the expensive kind of mistake.

So the first agent doesn't do any marketing. It collects information first

The fix is simpler than it looks. Fix the input before assigning any work. That fixed input is the brand brief.

So our team's first agent doesn't do any marketing. Its name is brand-brief-creator, and all it does is ask about brand information through conversation and produce a single standardized document.

Give it a website URL and it scrapes basic information from there to fill a draft, then asks a person about whatever is missing. The output is one file: outputs/brand-name/00_brand_brief.md.

There's a reason it's numbered 00. It means everything's starting point, zero before one.

There's one core design decision here. The orchestrator (the task dispatcher) checks whether a brief exists before starting any marketing work. If there isn't one, it stops everything else and calls brand-brief-creator first.

The code-like block below might look intimidating, but it's just "the gatekeeper checks admission tickets first" written as machine-readable rules.

Orchestrator rules:
  1. Check whether the brand brief exists first.
  2. If not present → call brand-brief-creator (pause all other work).
  3. If present → route P1 market research → P1.5 positioning → P2 growth ... in order.

This single gate is what holds up the quality of the entire team. Without input, every step above it stands on guesses, so this blocks the room where guesses could enter.

Without the brief the gate doesn't open. The orchestrator is a gatekeeper who checks admission tickets first.

Single source of truth: one page that many agents read together

The deeper reason a brief is the first button is that it is not "a memo only one person reads." It is a shared original that multiple agents read simultaneously.

This is called a Single Source of Truth, the principle that everyone looks at the same one place.

Spelled out concretely: who the target customer is, what the tone should be, who the competitors are, written in one place, and everyone looks at that one place.

Why does this matter? Suppose the market-research agent identifies the target as "restaurant owners in their 30s," but the content agent imagines the same brand as serving "solo renters in their 20s." Both work hard and yet the outputs don't align.

Because they're each drawing a different picture in their heads. This happens in human teams too, and in agent teams it's quieter, no one says "oh, we've been thinking about different targets." The brief eliminates that misalignment at the entrance. The actual flow becomes one-directional like this.

00_brand_brief  (single input)
      |
      +-> Market research  : uses target customer + competitive landscape fields as starting point
      +-> Positioning      : uses one-line positioning + 4 core messages to build 4P and messaging
      +-> Growth           : uses KPIs + conversion actions + current channels for media mix
      +-> Content          : uses tone + target insights for copy, SEO, and landing pages

Change one field upstream and everything downstream follows. Conversely, if the upstream is empty, every downstream agent fills that field with its own imagination. That is the seed of contradiction.

Water splitting from one document. Market research, positioning, growth, and content all receive the same source.

Anatomy of a good brief: 9 fact fields + 1 strategy field

So what goes on that one page? Our standard template has 10 sections. These aren't things to memorize and fill in, they are fields where an empty box means the marketing becomes guesswork.

Fields 1 through 9 are facts. They can be verified, and if wrong they can be corrected. What makes the real difference is field 10, especially the one-line positioning statement and the tone guide.

Here is a real example. When organizing a brief for a 30-year-old regional food wholesale distributor (anonymized as "Company A"), one line was locked into field 10:

"Not a flyer selling products, but a trade expert's information newsletter that's on the restaurant owner's side."

Why does that line matter? With it in place, the content agent won't write copy like "FLASH SALE!! LAST CHANCE!! BLOW-OUT!!"

Because the tone field also says "no oversaturated red-and-yellow clearance aesthetics; no clickbait expressions." A one-line positioning statement is a coordination device that makes every downstream agent speak in the same voice.

Without it, market research sounds measured, ad copy sounds loud, and they end up pulling in different directions.

9 fact fields to fill in; 1 strategy field (positioning and tone) that unifies the voice.

The hardest skill is confidently writing "unknown"

This is the thing I most wanted to say in this article. What separates a good brief from a dangerous one is not how complete it looks. It is how the unknown fields are handled.

Looking back at the Company A brief just mentioned, nearly half the fields turned out to be "TBD." Competitor names: TBD. Monthly budget: TBD. Target KPIs: TBD. Official logo, colors, and typefaces: undecided.

The temptation comes immediately. Empty fields look bad, so you want to fill them plausibly. "Competitors are big food wholesale chains." "Budget is around 3 million won a month." "Target is 30s to 50s." But the moment those go in, guesses pass themselves off as facts.

The next agent doesn't know they were guesses. If it says "competitors: large food wholesale chains," that gets treated as verified fact, and an entire competitive strategy is built on top of it. A precise false strategy stands on a false foundation.

So in our brief, unknown fields are honestly marked "TBD." Not just left blank, but left empty as a signal: "this is still unknown and therefore the next task".

The actual Company A brief ends with this: "Priority gaps: competitor names, budget, KPIs, and target data questions (currently TBD)." The list of unknowns becomes the list of next steps.

This is a principle that runs through this vault as a whole. Don't write estimates as if they are results. Mark unknowns as unknown, plant a flag there, and fill it in next time. The brief is simply the first document where that principle is applied.

Left: "TBD" fields flagged as future tasks. Right: guesses filling the boxes and passing as facts.

The same one page fans out into a flyer, a chat card, and a landing page

Calling the brief a single input isn't an abstraction. Seeing how one field unfolds into different deliverables makes it concrete.

Company A's brief has "4 core messages" in field 10. One of them is "the benchmark price, always in the same place." A sense of stability, reliability through consistency.

That one line wears different clothes in each channel. In the weekly print flyer it becomes a fixed benchmark product block that always appears in the same position. In a mobile KakaoTalk (a Korean messaging app) card it becomes the top price callout. On the landing page it becomes a trust section like "30 years, the same promise."

The key is that even though these three are different pieces of work, they come from the same single field. So the message doesn't waver as it crosses channels.

Without a brief, the flyer designer, the KakaoTalk operator, and the page planner would each independently define "what is the core of this brand?" Three subtly different brands would emerge.

The brief hands the same score to all three (or all three agents).

The same core message fans out into a flyer, a chat card, and a landing page. The root is always one field.

Even with a brief, half is still fog

After reading this far the brief might sound like a magic fix. Let me be honest about its limits.

First, a brief is only as good as what's been filled in. Starting with half of Company A's brief as "TBD" means the market-research and growth work above it also has to begin by "filling in the TBDs." The brief doesn't conjure information by magic. It just shows honestly where the holes are.

Second, the Company A brief itself started from a secondary source. It was restructured from an original memo handed over for a design outsource job into the standard 10-field format.

Values like "estimated target age 30–60" that went in through that process are inferences, not measured data. So they're marked "estimated" inside the brief itself.

This vault already contains an example of transcribing a handed-over memo and misrecording a fact in the process. So the same discipline is applied when filling the brief. Confirmed facts and estimated values are distinguished within each field.

Third, the brief must be a living document. When market research comes back with competitor names, that result can't just be recorded in a report and left there.

The "TBD" field in the brief must be reopened and updated so the next agent sees the current fact. Otherwise the single source of truth splits into two versions.

A brief is only as good as what's been filled in. The flags in the fog are not blanks to fill with lies, but questions to answer next.

The explanation ends here. Now fill out one page for your brand

You don't need a big agent team for this principle to apply. Whether outsourcing work or writing copy alone, filling out one page before starting a job saves half the trouble.

If you can only remember one thing, make it this:

The expensive mistake is not a wrong answer. It is a confident answer built on guessed input. Locking that input on one page is the brief.

Sources

The 10-section structure, gating rule, and "TBD" marking principle of the brand brief are drawn from this vault's actual deliverables and agent definitions. The 30-year food wholesale distributor cited as a case is anonymized as Company A; the original brief was itself written under a de-identification principle (excluding employee real names, internal systems, and sensitive financial figures). Values such as the estimated target age are inferences rather than measured data and are marked "estimated" inside the brief. The Single Source of Truth principle, primary-source verification, and distinguishing estimates from facts follow the operating rules of this vault (CLAUDE.md).

The client name is a pseudonym (Company A), and some figures are estimates rather than measured values. All company-identifying information in body text, tables, and captions has been de-identified.