Automate email responses with GPT-4o-mini and human review in Gmail

This n8n template demonstrates a “Human-in-the-Loop” workflow where AI automatically drafts replies to inbound emails, which are then reviewed and approved by a human before being sent. This powerful pattern ensures both the efficiency of AI and the quality assurance of human oversight.

Use cases are many: Streamline sales inquiry responses, manage first-level customer support, handle initial recruitment communications, or any business process that requires personalized yet consistent email replies.

Good to know

• At the time of writing, the cost per execution depends on your OpenAI API usage. This workflow uses a cost-effective model like gpt-4o-mini. See OpenAI Pricing for updated info.
• The AI’s knowledge base and persona are fully customizable within the Basic LLM Chain node’s prompt.

How it works

1. The Gmail Trigger node starts the workflow whenever a new email arrives in the specified inbox.
2. The Classify Potential Leads node uses AI to determine if the incoming email is a potential lead. If not, the workflow stops.
3. The Basic LLM Chain, powered by an OpenAI Chat Model, generates a draft reply based on a detailed system prompt and your internal knowledge base.
4. A Structured Output Parser is crucially used to force the AI’s output into a reliable JSON format ({"subject": "...", "body": "..."}), preventing errors in subsequent steps.
5. The Send for Review Gmail node sends the AI-generated draft to a human reviewer and pauses the workflow, waiting for a reply.
6. The IF node checks the reviewer’s reply for approval keywords (e.g., “approve”, “承認”).
7. If approved, the ✅ Send to Customer Gmail node sends the final email to the original customer.
8. If not approved, the reviewer’s feedback is treated as a revision request, and the workflow loops back to the Basic LLM Chain to generate a new draft incorporating the feedback.

How to use

• Gmail Trigger node: Configure with your own Gmail account credentials.
• Send for Review node: Replace the placeholder email [email protected] with the actual reviewer's email address.
• IF node: You can customize the approval keywords to match your team’s vocabulary.
• OpenAI Nodes: Ensure your OpenAI credentials are set up. You can select a different model if needed, but the prompt is optimized for models like GPT-4o mini.

Requirements

• An OpenAI account for the LLM.
• A Gmail account for receiving customer emails and for the review process.

Customising this workflow

• By modifying the prompt and knowledge base in the Basic LLM Chain, you can adapt this agent for various departments, such as technical support, HR, or public relations.
• The approval channel is not limited to Gmail. You can easily replace the review nodes with Slack or Microsoft Teams nodes to fit your internal communication tools.