When I’m building larger n8n workflows, there’s almost always a point where I want a human in the loop—either because accuracy matters at the end, or because you’re in an enterprise environment where “AI-generated content must be reviewed” is literally policy. In this guide, I walk through what human-in-the-loop means in n8n: the workflow pauses and waits for a person to approve or deny before it continues. I show the simplest starting point using Gmail approvals, where I generate something with an AI agent (like a YouTube title) and require approval before writing it into a spreadsheet. Then I step it up with approve/decline routing using an IF node, which is my preferred pattern because it gives you clean branching logic. I also cover wait limits (like a 45-minute timeout) so your workflow doesn’t hang forever. After that, I show the newer, much easier way to require approval before an AI agent can call a tool—like sending an email—using Telegram approvals as the gatekeeper. Finally, I cover the new chat-based human review inside n8n, and the confusing-but-important distinction between a general “send and wait” approval node versus the chat tool approval used specifically for agent tool calls. I wrap with a real-world challenge: auto-categorize support tickets with AI, but require human review on low confidence or escalation keywords before routing/responding.