---
title: Brand and tone
summary: 'Keep AI-generated communication aligned with trust, clarity, and voice.'
agent_summary: >
  This page explains the brand and tone guardrail, the decision it governs, what
  drift looks like, and how the system responds.
canonical_url: /docs/guardrails/brand-and-tone
page_type: guardrail
related_resources:
  - /resources/guardrails/brand-tone.v1.json
related_schemas:
  - /schemas/guardrail.schema.json
last_reviewed: '2026-04-09'
---
# Brand and tone

Keep AI-generated communication aligned with trust, clarity, and voice.

> Agent summary: This page explains the brand and tone guardrail, the decision it governs, what drift looks like, and how the system responds.


## Headings
- ## Why this matters
- ## What decision is being governed
- ## What good judgment looks like
- ## What drift looks like
- ## Example in practice
- ## Boundaries
- ## How JudgmentKit responds
- ## Ownership and review
- ## Technical reference
- ## Related pages

## Why this matters

AI can stay technically correct and still feel wrong. It can sound too certain, too casual, too urgent, or simply unlike the product people trust.

## What decision is being governed

This guardrail governs how the system speaks when it generates or rewrites language for users.

## What good judgment looks like

- clear, grounded language
- confidence that matches evidence
- reassurance in support contexts
- concise product copy that still sounds human

## What drift looks like

1. Unsupported certainty
2. Deceptive urgency
3. Voice inconsistency across surfaces
4. Over-simplification in sensitive flows

## Example in practice

The launch example shows a support assistant turning a billing response into an urgent, coercive instruction. The fix is not just softer language. It is confidence that matches the evidence and next steps that reduce anxiety instead of increasing it.

## Boundaries

Allowed variation includes reading-level changes, surface-aware concision, and small tone shifts between support and builder contexts.

Hard stops include deceptive urgency, unsupported guarantees, fear-based persuasion, and humor in sensitive flows.

## How JudgmentKit responds

Low drift gets normalized and logged. Medium drift gets rewritten and reviewed. High drift gets blocked or escalated because the main risk is trust loss, not only cosmetic inconsistency.

## Ownership and review

Design Systems owns the decision. Brand owns the risk. Platform Engineering owns the runtime behavior that turns drift signals into action.

## Technical reference

- Resource: `/resources/guardrails/brand-tone.v1.json`
- Schema: `/schemas/guardrail.schema.json`

## Related pages

- /docs/workflows/support-assistant
- /docs/examples/brand-tone-support-coercion
- /docs/guardrails/safety-and-privacy

## Related pages
- /docs/workflows/support-assistant
- /docs/examples/brand-tone-support-coercion
- /docs/guardrails/safety-and-privacy

## Related resources
- /resources/guardrails/brand-tone.v1.json

## Related schemas
- /schemas/guardrail.schema.json