n8n Gmail Automation: OAuth2 Setup, AI Auto-Replies & Free Templates

⏱️ 18 min read · 📥 2 free importable templates included

Last year I built a client an invoice-reminder automation. It worked for exactly seven days, then died silently — no error, no log, just an inbox that stopped responding. The cause wasn't the workflow. It was a Google Cloud setting with a seven-day fuse that no guide mentioned. That setting is section 2 of this article, and it's the difference between an automation that survives the week and one you rebuild every Monday.

1. What You Can Automate

Gmail is one of the most useful integrations in n8n, because email touches support, sales, invoices, notifications, and internal operations — almost every business process eventually routes through an inbox. The map, in the order you'll actually need it:

2. The 7-Day Token Trap (Fix This First)

Here's the bug nobody warns you about, and the one that cost my client a silent week: Google Cloud projects default to "Testing" mode, and in Testing mode, Gmail refresh tokens expire after roughly 7 days. No error, no warning — your automation simply stops connecting a week in.

The fix is usually permanent, but read the fine print below before you assume it's bulletproof:

3. Google Cloud OAuth Setup (The Part Everyone Gets Wrong)

The full path, in order. Skip a step and you'll meet one of the errors in section 11:

Follow the principle of least privilege — only request the scopes your workflows actually use:

Building everything in this guide (trigger + auto-reply + labeling) needs all three. If you only need the daily-digest template, gmail.readonly alone is enough — don't request gmail.modify or gmail.send if nothing in your workflow uses them.

4. Connect n8n (Credential)

In n8n: Credentials → Add credential → Gmail OAuth2. Paste the Client ID and Client Secret. n8n will display an OAuth redirect/callback URL directly inside this credential screen — copy that exact URL into your Google Cloud OAuth client's "Authorized redirect URIs" field. This is the setting that actually controls whether the OAuth handshake succeeds; it is easy to confuse with other URL settings, so here's the difference:

If you're self-hosting behind a reverse proxy with HTTPS, make sure WEBHOOK_URL reflects your real public HTTPS domain — our Docker guide covers the full reverse-proxy setup.

One credential, two nodes: the Gmail Trigger (incoming) and the Gmail node (outgoing) both select it. Connect once and both work.

5. Reading: The Trigger & Filter Queries

The Gmail Trigger works by polling — it checks your inbox on an interval you configure, it does not push instantly like a webhook. Its most useful setting is the q filter: a Gmail search query that decides which emails even reach your workflow. The operators worth memorizing:

Combine them like Gmail search: from:[email protected] has:attachment is:unread. The trigger's output is the full email: $json.subject, $json.textPlain (body), $json.id (the message ID — what the Reply operation needs), and binary attachments when you enable download.

6. Sending: Text, HTML, Drafts & Attachments

The Gmail node handles the outgoing side. These four operations cover most workflows you'll build:

• Send — delivers immediately. Fields: To, Subject, Message.
• Reply — answers the email that triggered the workflow (needs the email's id).
• Draft — creates the email in your drafts without sending. The approval-friendly choice: pair it with our Human-in-the-Loop guide and nothing gets sent until a human reads it.
• Mark as Read / Add Labels — inbox housekeeping after processing, and the key to preventing reply loops (section 8).

Three details that elevate an automation from "works" to "looks professional":

• HTML emails: choose the HTML type and use basic tags (<b>, <br>, inline styles). Skip complex CSS — Gmail strips it.
• Attachments: in the node's Options, add attachments from binary data of previous nodes or from URLs.
• Reply-to: set a different reply address than the sender when the reply should land elsewhere.

7. The AI Email Assistant

The modern upgrade: attach the Gmail tool to an AI Agent, and the agent reads, drafts, and sends as part of its reasoning — you write the policy, the agent writes the mail. Two rules from running this in production:

• Never send without approval. The agent's Gmail tool belongs behind an approval checkpoint — our HITL guide wires the review step in one minute. An agent that can email customers unsupervised is a PR incident waiting.
• Give it context, not just instructions. The agent drafts better when it sees the customer's history — memory from our AI Agent Memory guide plus your RAG knowledge base from the RAG tutorial.

The pragmatic setup most people actually need: Trigger → Agent drafts a reply → Gmail node creates a Draft → you review and hit send. Zero risk, full speed.

8. Before You Activate

Any workflow that sends or replies automatically deserves a five-minute safety check before you flip it on:

• Test with a throwaway account first. Send yourself emails from a second Gmail address and watch the workflow run manually before activating it against real contacts.
• Never let an auto-responder reply to itself. Exclude your own sending address and generic no-reply@ senders from the trigger query.
• Mark every processed email. Apply a label (e.g. n8n-processed) right after acting on an email, and exclude that label from future polls — otherwise the same email can be replied to on every single poll cycle.
• Don't reply to mailing lists or bulk senders. Add exclusions for known newsletter/marketing senders if your inbox receives them.
• Check your daily send quota (section 10) before pointing an automation at a busy inbox.

9. Free: 2 Importable Templates

Both templates use real n8n node structures. Import, pick your credential, adjust the filters, test manually, then activate.

Template 1 — Smart Auto-Responder (with loop protection)

Watches for a specific sender and replies with a tailored message. Three design choices make it safer than a basic version: it uses the Reply operation (answers the original message ID directly — no address-parsing to break), the q filter excludes anything already labeled as processed, and a final node applies that label so the same email is never answered twice.

{
  "name": "Gmail Smart Auto-Responder",
  "nodes": [
    {
      "parameters": {
        "pollTimes": {
          "item": [
            {
              "mode": "everyMinute"
            }
          ]
        },
        "filters": {
          "includeSpamTrash": false,
          "includeDrafts": false,
          "q": "from:[email protected] is:unread -label:n8n-processed"
        },
        "options": {}
      },
      "id": "trigger",
      "name": "Gmail Trigger",
      "type": "n8n-nodes-base.gmailTrigger",
      "typeVersion": 1.3,
      "position": [250, 300],
      "credentials": {
        "gmailOAuth2": {
          "id": "YOUR_GMAIL_OAUTH_ID"
        }
      }
    },
    {
      "parameters": {
        "resource": "message",
        "operation": "reply",
        "messageId": "={{ $json.id }}",
        "emailType": "text",
        "message": "Hi,\n\nThanks for your message. I will get back to you within one business day.\n\nBest,\nWilliam",
        "options": {}
      },
      "id": "reply",
      "name": "Reply to Sender",
      "type": "n8n-nodes-base.gmail",
      "typeVersion": 1,
      "position": [450, 300],
      "credentials": {
        "gmailOAuth2": {
          "id": "YOUR_GMAIL_OAUTH_ID"
        }
      }
    },
    {
      "parameters": {
        "resource": "message",
        "operation": "addLabels",
        "messageId": "={{ $json.id }}",
        "labelIds": ["YOUR_N8N_PROCESSED_LABEL_ID"]
      },
      "id": "add-label",
      "name": "Mark as Processed",
      "type": "n8n-nodes-base.gmail",
      "typeVersion": 1,
      "position": [650, 300],
      "credentials": {
        "gmailOAuth2": {
          "id": "YOUR_GMAIL_OAUTH_ID"
        }
      }
    }
  ],
  "connections": {
    "Gmail Trigger": {
      "main": [
        [
          {
            "node": "Reply to Sender",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Reply to Sender": {
      "main": [
        [
          {
            "node": "Mark as Processed",
            "type": "main",
            "index": 0
          }
        ]
      ]
    }
  },
  "active": false,
  "settings": {
    "executionOrder": "v1",
    "saveManualExecutions": true
  }
}Copy

Template 2 — Daily Unread Email Counter (sent to Telegram)

What this actually does: every morning at 8, it counts unread emails from roughly the last day and sends the count as a message to Telegram. It is a counter, not an AI summary — it does not read or summarize email content. If you want real summaries of subject lines and senders, add an AI node between "Get Unread Emails" and "Build Message" (see the note below the code).

{
  "name": "Gmail Daily Unread Counter to Telegram",
  "nodes": [
    {
      "parameters": {
        "rule": {
          "interval": [
            {
              "field": "cronExpression",
              "expression": "0 8 * * *"
            }
          ]
        }
      },
      "id": "schedule",
      "name": "Every Day 8:00 AM",
      "type": "n8n-nodes-base.scheduleTrigger",
      "typeVersion": 1.2,
      "position": [250, 300]
    },
    {
      "parameters": {
        "resource": "message",
        "operation": "getAll",
        "returnAll": true,
        "additionalFields": {
          "q": "is:unread newer_than:1d"
        }
      },
      "id": "get-unread",
      "name": "Get Unread Emails",
      "type": "n8n-nodes-base.gmail",
      "typeVersion": 1,
      "position": [450, 300],
      "credentials": {
        "gmailOAuth2": {
          "id": "YOUR_GMAIL_OAUTH_ID"
        }
      }
    },
    {
      "parameters": {
        "jsCode": "const emails = $input.all();\n\n// This counts unread emails — it does NOT summarize content.\n// To add real summaries, insert an AI node before this step\n// and pass subject/sender fields into the prompt.\nconst count = emails.length;\nconst summary = `📬 Morning check: ${count} unread email${count === 1 ? '' : 's'} from roughly the last 24 hours.`;\n\nreturn [{ json: { count, summary } }];"
      },
      "id": "summarize",
      "name": "Build Message",
      "type": "n8n-nodes-base.code",
      "typeVersion": 2,
      "position": [650, 300]
    },
    {
      "parameters": {
        "chatId": "YOUR_TELEGRAM_CHAT_ID",
        "text": "={{ $json.summary }}"
      },
      "id": "telegram",
      "name": "Telegram Digest",
      "type": "n8n-nodes-base.telegram",
      "typeVersion": 1.2,
      "position": [850, 300],
      "credentials": {
        "telegramApi": {
          "id": "YOUR_TELEGRAM_CREDENTIAL_ID"
        }
      }
    }
  ],
  "connections": {
    "Every Day 8:00 AM": {
      "main": [
        [
          {
            "node": "Get Unread Emails",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Get Unread Emails": {
      "main": [
        [
          {
            "node": "Build Message",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Build Message": {
      "main": [
        [
          {
            "node": "Telegram Digest",
            "type": "main",
            "index": 0
          }
        ]
      ]
    }
  },
  "active": false,
  "settings": {
    "executionOrder": "v1",
    "saveManualExecutions": true
  }
}Copy

• One credential for both templates — YOUR_GMAIL_OAUTH_ID, the same credential from section 4.
• Verify your output shape once: run the trigger manually (Execute workflow) and inspect the JSON before wiring downstream nodes — field names vary between the "simple" mode and raw payloads, and between n8n versions.
• Want a real AI summary instead of a count? Insert an AI node (OpenAI/Anthropic) between "Get Unread Emails" and "Build Message" that receives subjects and senders and returns 2–3 bullet points — be mindful of sending email content to a third-party API if it contains sensitive data.
• Extend the digest to Gmail instead of Telegram: swap the Telegram node for a Gmail Send node — same structure, one node change.
• More ready-to-import blueprints live in our n8n Workflow Templates Library.

10. Rate Limits & Quotas

The numbers that decide whether your automation is polite or blocked:

11. The 10 Most Common Errors & Fixes

12. Frequently Asked Questions

Sources & References

• n8n Documentation — Gmail node
• n8n Documentation — Gmail Trigger node
• Google — OAuth token expiration (testing mode behavior)
• Google — Gmail sending limits
• Google — Gmail search operators

Explore More Guides

Ahmed Ayari— TriggerWorkflow The seven-day story in the intro is not a metaphor — it's the invoice-reminder project that taught me Google's Testing mode the hard way, and section 2 exists so you never repeat it. Both templates in section 9 were tested on a live Gmail account with n8n's current stable release, and every error in section 11 is one I've hit, fixed, and written down. Read more guides by Mr.Ayari