On this page
Built-in policies catch the obvious stuff. Guidelines let you write custom rules and enforce them with AI.
A dating app that prohibits asking for last names. A marketplace that blocks external payment links. A gaming community where "I'll kill you" is trash talk, not a threat.
Getting started
Describe the rule, and decide whether it should flag content or just run as analysis.
If you've added context for your project, the AI factors that in automatically.
Include conversationId and authorId in your requests so the AI can evaluate messages in context.
const result = await moderationApi.content.submit({
content: {
type: "text",
text: message.text,
},
conversationId: thread.id,
authorId: message.author.id
});Writing good guidelines
Good guidelines are specific and verifiable.
- "Must not ask for personal contact information" beats "Be respectful of privacy."
- "Must not list salary below minimum wage" beats "Ensure fair compensation."
- "Must include trigger warning before discussing self-harm" beats "Be mindful of others."
Prohibitions are easier to evaluate than vague principles.
Built for moderation
Guidelines aren't just an LLM prompt.
Context. Every request includes conversation history, author reputation, and platform context, automatically.
Feedback loops. When something misfires, you get the tooling to catch it, understand why, and iterate. Not just a prompt you tweak and hope.
Maintenance. Models change. Best practices change. We keep things updated so you don't have to.
No engineering required. Your trust & safety team can add and update guidelines directly in the dashboard. No deploys, no prompt wrangling.
Try it out
Head to your dashboard and add your first guideline. Start with flagging disabled to observe performance before enforcing.
Not sure how to write effective guidelines? Reach out, we're happy to help.

About the author
Christopher Dengsø
CEO, Moderation API
Christopher builds Moderation API and writes about content moderation, trust and safety, and the product.



