Build your first reviewable AI workflow with n8n
Goal
Build a basic workflow that accepts text, turns it into a structured summary, waits for human approval, and preserves the input and output of every stage.
Prerequisites
- Access to n8n Cloud or a self-hosted n8n instance.
- Test text that contains no sensitive information.
- A configured model or API credential stored in n8n credential management rather than inside a node.
Steps
- Create a workflow with a manual trigger so external systems do not add variables during the first test.
- Add input fields for test text and the expected output structure.
- Add a processing node whose prompt is limited to extracting topics, action items, and facts that still need confirmation.
- Add a condition after the model node; send output missing any required field to a failure branch.
- Add a human-approval stage and prevent the workflow from writing to external systems before approval.
- Add the final output node, run three samples of different lengths, and inspect each node's execution data.
Expected result
The workflow consistently returns the same structure, the failure branch identifies missing fields, and the execution record shows the stage where an error occurred.
Troubleshooting
- If the output structure varies, use structured output or a template node for deterministic formatting.
- If a node fails, inspect that node's input variables instead of rerunning the whole workflow immediately.
- Before enabling a webhook or schedule, configure timeouts, retries, and notifications.