Brand voice rules that your content AI can't accidentally violate

Most content teams have a brand voice guide. It's in Notion, or Google Docs, or a carefully formatted PDF. It says things like: we use "we," not "I." We don't use corporate jargon. We write in second person. We never say "leverage." We acknowledge uncertainty; we don't project false confidence.

Most AI-generated content doesn't load this guide. The content writer uses Claude or ChatGPT, writes a first draft, reviews it, fixes the brand violations, publishes. The violations were caught because a human reviewed. The AI had no access to the voice guide.

This works fine at low volume. It breaks down when content velocity increases, when more people on the team are generating content with AI assistance, or when the same brand violations appear in the same places every time because nobody operationalized the fix.

The gap between the guide and the session

The root problem: a brand voice guide is a document. An AI content session loads whatever context the user included. If the user didn't include the brand guide — because they were in a hurry, because they forgot, because the guide is in a system nobody thought to attach — the guide might as well not exist.

This isn't a people problem. No reasonable workflow requires every content team member to manually include the brand guide in every AI session. The guide needs to load automatically, not be remembered to be included.

Brand rules as engine rules

In Memory Stack, brand voice rules are stored the same way operational rules are: as memories, tagged for the agents and users that should load them. A content writer using Claude or ChatGPT with Memory Stack connected starts every session with the brand rules already loaded. No manual inclusion. No "did you remember to attach the style guide."

The rules can be as granular as the content team needs:

• "Use 'we' throughout. Never 'I' or the company name as subject in running prose."
• "Avoid: leverage, utilize, synergy, robust, seamless, game-changing."
• "Acknowledge what we don't know. Don't project certainty we don't have."
• "Headlines in sentence case. No punctuation at end of headline."
• "Technical posts: include a concrete example before the abstract explanation, not after."

These aren't prompting instructions. They're rules that load at session start and stay in context for the session.

When the guide changes

Brand guides evolve. A rebrand, a new tone principle, a specific word that's been banned after a campaign went the wrong direction. Under the standard workflow — guide in a doc, included when someone remembers — updates propagate when whoever shares the doc shares the new version, and until then, people work from the old one.

Under the engine rules model: update the memory. The next session that loads it gets the updated rule. Every content writer who's connected to Memory Stack is on the same version of the brand guide, on the next session after the update.

This is especially relevant when a brand principle is being actively enforced. A new competitor landed and legal has a sensitivity about certain phrasing. Marketing decided "AI-powered" is overused and has made it a banned phrase. These need to propagate fast. Memory Stack propagates them at session-init speed.

Governance for content at scale

At small team scale, a brand voice guide review by a skilled editor catches violations reliably. At larger content scale — multiple writers, AI-assisted drafts, content moving through multiple stages of production — the probability that every piece gets a full brand review before publication drops.

Engine rules don't replace editorial review. They reduce the error rate at the generation stage, so review time goes to higher-value problems than "they wrote 'leverage' again."

The audit trail is also useful for content teams. When a brand-sensitive piece of content is questioned, the session log shows which brand rules were loaded when it was generated. The content team can verify that the rules were current and present, or identify that a rule was missing and take it as a capture gap to close.

The honest version

Rules loaded in context are recommendations to the AI, not hard constraints. A sufficiently long or complex content session can still produce output that drifts from the rules. The rules are reliable guardrails at session start; they're not a guarantee across a 3,000-word document produced in one session.

Review is still necessary. The bar it achieves: brand violations are less frequent and more consistent. The violations that do appear are more likely to be genuine edge cases, not systematic drift from a guide that wasn't loaded.

Also: the quality of the rules matters. A brand guide that's vague ("write with warmth") produces vague rule enforcement. Concrete rules ("use a concrete example before the abstract explanation") produce consistent, testable behavior.

Brand voice is the accumulation of intentional decisions your content team made about how to communicate. Those decisions should be in every AI content session by default — not when someone remembers to include them.