AI reply drafts for Gmail: free n8n workflow template

By the serverserv team · 3 min read · Updated 28 October 2026

This workflow reads each new email in your inbox, asks an AI model to write a reply in your tone, and saves it as a Gmail draft in the same thread. You open Gmail, check the draft, fix anything marked [CHECK] and press send.

Download the template

How it works

1. New email checks Gmail every minute for unread messages in your inbox.
2. Skip automated senders ignores no-reply addresses, notifications and bounce messages.
3. Write reply sends the sender, subject and the first 6,000 characters of the email to the AI model, with instructions to keep it short and never guess facts.
4. Save draft reply creates a draft addressed to the sender, in the original thread.

Nodes used: Gmail Trigger, If, AI Agent, OpenAI Chat Model, Gmail.

What you need

• A Gmail or Google Workspace account
• An OpenAI API key, or another chat model connected to the AI Agent

Set it up

1. Import the workflow. Connect Gmail in New email and Save draft reply (use the same credential).
2. Open OpenAI Chat Model, add your API key and pick a model. A small, low-cost model is usually enough for email.
3. Edit the system message in Write reply: add what your business does, your opening hours, your sign-off and anything it should never promise.
4. Activate the workflow and send yourself a test email from another address.

New to importing? In n8n, open a new workflow, then either paste the copied JSON onto the canvas or use the menu (the three dots) and choose Import from file. Credentials are never included, so you connect your own accounts after importing.

Ways to customise it

• Add a Gmail label filter so only one label (for example Support) gets drafts.
• Give the agent a tool that reads your FAQ or a Google Doc so it can answer common questions properly.
• Replace the OpenAI node with Anthropic, Gemini or a local Ollama model.

Security notes

Emails are untrusted input. The system message tells the model to answer only from the email and to mark gaps with [CHECK], and the workflow only ever creates drafts, never sends. Keep it that way: a model that can send email on its own can be tricked by a cleverly written message.

Where to run it

Templates with webhooks or triggers need n8n running all the time and reachable over HTTPS. You can run n8n on your own computer to try it, but for real use it belongs on a server. serverserv n8n Hosting gives you n8n pre-installed on your own KVM server with HTTPS, from $14.99 a month with no execution limits, and we will set up your first workflow for you, this one included.

Questions

Sources

• n8n Docs: Export and import workflows
• n8n Docs: Self-hosting n8n