Send validated AI output from n8n

Use a model step to produce content, then a separate HTTP Request step to send it. Start with the n8n HTTP guide.

Define the model output

Require an object matching this JSON Schema:

{
  "type": "object",
  "properties": { "message": { "type": "string", "minLength": 1, "maxLength": 1000 } },
  "required": ["message"],
  "additionalProperties": false
}

Validate the result before delivery. Use a template with {{message}}; map the parsed object’s message into data.message. Keep the template, destination, credentials, and stable operation ID in workflow configuration, outside model output.

Route the outcome

Add a review/approval step before sends when required by your workflow. Handle invalid model output without calling the API. On a successful send, save the returned message coordinates; on failure, record the request ID and error code. A workflow retry must reuse the same approved content and operation ID. Do not regenerate text and reuse the old key, or silently generate a new key after uncertain delivery. Agent design · HTTP request settings

HTTP workflows

Agent tools

Preview integrations

Send validated AI output from n8n

Define the model output

Route the outcome

​Define the model output

​Route the outcome

Define the model output

Route the outcome