Use case
AI content moderation workflow
Updated September 17, 2026
Moderation needs consistent decisions and a record of why each one was made. A structured verdict from Claude, a Switch with three outcomes, and a stored run for every post give you both.
How the workflow runs
- The trigger receives the post text and author.
- A review agent applies your policy in its system prompt and returns
verdictandreasons. - A Switch fires
allow,block, ordefault, which queues the post for a human. - Each branch shapes the outcome with a Set node; Merge and Output return it.
A verdict you can branch on
{
"type": "object",
"properties": {
"verdict": { "type": "string", "enum": ["allow", "review", "block"] },
"reasons": { "type": "array", "items": { "type": "string" } }
},
"required": ["verdict", "reasons"]
}Uncertain cases fall through to default, so anything the model is unsure about reaches a person rather than being allowed.
An audit trail for free
Every run stores the exact graph that ran, the prompt the agent received, and the verdict it returned. When someone appeals a decision, open that run and see why it was made.
Keep reading
- AI support ticket triageBuild an AI support ticket triage workflow: Claude classifies urgency and topic with structured output, urgent tickets escalate, and the rest get a drafted reply.
- Agent node: Claude with tools and structured outputThe Agent node calls Claude with your prompt, can call attached HTTP tools in a loop, and returns text plus schema-validated structured output for branching.
- Switch node: multi-way branchingThe Switch node routes a workflow down one of many branches: each case has its own port and expression, with a default port when none match.
Build it on the canvas
Create a free account and run your first workflow in the browser, or run Novalink on your own machine.