Practical governance and authentic ownership

A badge describes a process. It does not establish that the process was adequate. Choose controls in proportion to the audience, purpose, and consequences.

Distinguish the work

Practice	What actually happens	What it does not imply
Human authorship	A person composes the message without generative drafting.	Accuracy, care, or superior quality.
AI-assisted editing	AI supports human-led composition.	That no AI wording remains.
Human-in-the-loop review	Someone evaluates a particular AI draft and can correct or reject it.	Explicit approval unless the person accepts the final version.
Human approval and ownership	An authorized person approves that final version and owns its claims and commitments.	Independent audit or fact-checking.
Automated policy validation	Defined controls evaluate output or constrain generation.	That a human read the message or that every error can be caught.
Fully automated communication	The system sends without individual pre-send human review.	Absence of governance, or absence of accountable humans.

HITL is meaningful only when reviewers have the ability, context, time, and authority to evaluate, correct, or reject the message. A rubber-stamp approval is not meaningful oversight. If output changes after approval, it needs review again before claiming approval of the final version.

Minimum Governed AI disclosure

Name the controls actually applied. Also state whether every message gets human review and whether sending is automated. Name a responsible team or person and a contact channel. Keep the short disclosure with the message; a link can provide additional detail.

Example: “AI generated and sent this response automatically. Controls: retrieval limited to approved support articles, personal-data filtering, and escalation of low-confidence answers. No individual pre-send human review. Accountable team: Customer Support, support@example.com.”

For an internal workflow record, document its purpose, control owner, allowed inputs and recipients, checks performed, failure behavior, escalation route, version, and review date. Evaluate those controls against actual failure cases. Changes to sources, prompts, models, or distribution can require re-evaluation. Do not claim independent auditing, verification, or fact-checking unless those specific processes occurred.

Examples by context

Context	Suitable disclosure	Appropriate practice
Business correspondence	AI Assisted for a human-led proposal edited with AI.	Sender checks meaning, commitments, recipient, and confidentiality.
Marketing content	Human Reviewed for AI-drafted copy approved by an editor.	Editor checks claims, tone, rights, and the final published version.
Automated notifications	AI Automated, with the accountable operations team and contact.	Describe any applied controls. Make it easy to reach a human.
Executive communications	Human Reviewed when AI drafted and the executive approved the final text.	Confirm the named sender's intended meaning and commitments; delegation should be explicit.
Higher-risk professional communications	A precise account of review and controls, alongside the primary signal.	Follow the relevant professional and organizational requirements. A badge cannot replace qualified review or required disclosures.

No scoring system belongs here. Human Written is not “more trustworthy” than Governed AI. Honest disclosure and appropriate ownership matter across all five practices.