The Brand Constitution: How to Automate Brand Governance

The Hidden Editing Tax

A mid-market brand used an agent to generate 500 product descriptions, finishing the first draft in under an hour. See what UK brand and marketing teams are learning about AI governance for the research behind this.

A junior marketing manager was asked to review the copy. By the third description, problems were obvious.

‘Delve into a world of comfort.’

‘Unleash the power of sustainable living.’

What should have been a quick check became a week of manual editing. The manager fixed the same errors on a loop: stripping clichés, tightening tone, and adding back mandatory details.

The agent hadn’t saved work; it had just shifted it to expensive human oversight.

The Prompt Mirage

The team believed a prompt could guarantee on-brand content. Managers saw coherent text and assumed the job was done. It failed for two reasons.

• Brand knowledge is implicit. The real voice of a brand lives in people’s heads. It is shaped by unwritten rules, past decisions, and judgement built over time. None of this exists in a prompt unless made explicit.
• Scale makes this worse. The same mistake across 500 descriptions becomes an editing tax. Humans stop editing and start policing output.

The Fix: Treat Brand Voice as a System

A Brand Constitution is a set of machine-readable rules that define how the brand speaks. A second agent enforces them.

1. Brand Constitution Schema

2. The Critic’s Scorecard

Operational Setup

1. Codify the Brand Voice

Gather your top 20 best-performing emails, product pages, or articles. Feed them into a model and extract the rules that already exist but are undocumented.

Prompt: “I am pasting 20 examples of our best writing. Analyse the tone, sentence structure, and vocabulary. Output a structured ‘Brand Constitution’ with the following fields:

1. Five adjectives that describe our voice
2. Five forbidden phrases we never use
3. Our rules for sentence length and structure.”

Refine the output with your CMO or Brand Lead. Instinct becomes syntax.

2. Build the Critic Agent

Its role is narrow. Do not let it write.

System Prompt:

Role: You are the Brand Critic. You do not write. You only grade.

Input: A draft product description.

Context: [PASTE BRAND CONSTITUTION HERE]

Task: Check the draft against the Constitution.

Output: Provide a structured response with the following fields: status (either “PASS” or “FAIL”), violations (a list of specific rules broken), and feedback (instructions for the writer).

3. Connect the Loop

Use an orchestration tool (like Make or Zapier) to wire the agents together.

1. The writer agent produces a draft.
2. The critic checks it.
3. If PASS: It moves to human review.
4. If FAIL: The structured feedback is sent back to the writer agent.
5. The writer produces a revised version immediately.
6. The critic checks it again.

Only drafts that fail twice are escalated.

4. Review Exceptions, Not Drafts

At that point, editors stop reviewing drafts. They review decisions. On a weekly cadence, they scan critic failures and spot-check passes. When patterns emerge, they update the Constitution. Humans stop fixing copy. They maintain the system that governs it.

Executive Takeaway

By separating creation from critique and by building a self-healing loop, agents don’t just generate volume. Value comes from the loop itself. Humans stop fixing repetitive errors and start improving the rules of the game. Speed increases while the brand stays protected.