PromptMarket

Prompt

Content moderation classifier

Flag a message against a short policy, and quote the part that triggered it.

Moderationintermediatemoderationpolicyclassification

When to use

You need a first pass that routes illegal, abusive, or policy-breaking text to a person or a block.

Prompt

You moderate one message against the policy. Do not reply to the user.

Policy categories:
{{categories}}

Return JSON:
{
"decision": "allow" | "review" | "block",
"category": string | null,
"quote": string | null
}

Choose `allow` when no category's rule matches. Choose `review` when a person should see it. Choose `block` only for categories the policy marks as block. `quote` is the triggering span, or null when the decision is allow.

Message:
{{input}}

Variables

  • {{categories}}
  • {{input}}

Example

Input

I will post your home address unless you refund me today.

Output

{
  "decision": "review",
  "category": "threat",
  "quote": "I will post your home address"
}

Why it works

The categories are a policy you wrote. The quote shows the trigger, and allow is the default when nothing matches.

Common mistakes

  • A category so broad that ordinary complaints are flagged.
  • Flagging without a quote, which nobody can review.
  • Using the classifier as the only control on a high-cost action.

Related concepts

Related prompts