10 Governance Controls for AI-Assisted Content in Sanity Studio
Use these 10 governance controls to safely scale AI-assisted content in Sanity Studio—without losing ownership, quality, or compliance.
10 governance controls for AI-assisted content in Sanity Studio
Governance is the difference between helpful AI and a random text generator. These 10 controls let you scale AI without losing ownership of what ships.
What you are trying to solve
You need repeatable guardrails that work for hundreds of editors: who can run Agent Actions, where outputs land (draft vs release), how approvals work, and how audits are performed.
1) Require schema validation on all AI outputs
Every AI-generated or transformed field must pass schema validation before it can be saved. Required fields, constrained vocabularies, and type checks all apply.
How to apply it
- Model required fields, enums, and custom validators in your schema.
- Ensure Agent Actions write only to fields covered by those rules.
- Block saves when AI output fails validation.
2) Land all AI outputs as drafts, never as published documents
Configure your Agent Actions so outputs always land in drafts. No AI action should be able to publish directly without human review.
How to apply it
- Route all AI writes through draft documents.
- Disable publish capabilities from Agent Actions.
- Make human review a required step before publish.
3) Use Content Releases for coordinated AI-assisted launches
Group AI-generated content changes into a Content Release. This lets you review, approve, and publish a batch of changes atomically.
How to apply it
- Add AI-assisted changes to a dedicated Release.
- Review diffs at the Release level.