Smart Gmail auto-labeler with Gemini AI & sender history

Description

n8n Gmail AI Auto-Labeler

> An intelligent n8n workflow that automatically classifies and labels Gmail emails using Google Gemini AI, keeping your inbox organized with zero manual effort.

This workflow uses AI-powered classification to analyze email content, learn from sender patterns, and automatically apply appropriate labels while archiving processed emails.

How It Works

Trigger: The workflow runs automatically every minute to check for new unread emails (or manually for bulk processing).
Check for Existing Labels: Before processing, it verifies if the email already has an AI-assigned label to avoid duplicate processing.
AI Classification: If unlabeled, the AI agent analyzes the email using:
Sender History Tool - Fetches up to 10 previous emails from the same sender to identify patterns
80% Majority Rule - If 80%+ of sender's past emails have the same label, strongly prefers that category
Label Examples Tool - When uncertain, compares the email with existing examples from suspected categories
Smart Decision: The AI returns a structured JSON response:
{
"label": "Category Name"
}
Or "None" if no category fits.
Apply & Archive:
Label Applied → The workflow adds the appropriate Gmail label to the thread.
Auto-Archive → Removes the email from INBOX (archives it) to maintain zero-inbox.
Loop: Processes the next email in the batch, ensuring all new emails are classified.

Requirements

Gmail OAuth2 Credentials** - Connected Gmail account with API access.
Google Gemini API Key** - Get it here
Free tier: 15 requests/minute
Gmail Labels** - Must be created in Gmail exactly as listed:
Meetings
Income
Inquiries
Notify / Verify
Expenses
Orders / Deliveries
Trash Likely

How to Use

Import the Workflow:
Copy the provided JSON file.
In your n8n instance → click Import Workflow → select the JSON file.

Create Gmail Labels:
Open Gmail → Settings → Labels → Create new labels.
Use the exact names listed above (case-sensitive).

Get Your Label IDs:
In the workflow, click "When clicking 'Execute workflow'" manual trigger.
Execute the "Get Labels Info" node only.
Copy each label's ID (format: Label_1234567890123456789).

Update Code Nodes with Your Label IDs:

Node 1: "Check Label Existence"
const labelMap = {
"Label_YOUR_ID_HERE": "Meetings",
"Label_YOUR_ID_HERE": "Inquiries",
"Label_YOUR_ID_HERE": "Notify / Verify",
"Label_YOUR_ID_HERE": "Expenses",
"Label_YOUR_ID_HERE": "Orders / Deliveries",
"Label_YOUR_ID_HERE": "Trash Likely"
};

Node 2: "Convert Label to Label ID"
const labelToId = {
"Meetings": "Label_YOUR_ID_HERE",
"Inquiries": "Label_YOUR_ID_HERE",
"Notify / Verify": "Label_YOUR_ID_HERE",
"Expenses": "Label_YOUR_ID_HERE",
"Orders / Deliveries": "Label_YOUR_ID_HERE",
"Trash Likely": "Label_YOUR_ID_HERE"
};

Set Up Credentials:
Gmail OAuth2 → Authorize your Gmail account in n8n.
Google Gemini API → Add your API key in n8n credentials.

Test the Workflow:
Send yourself test emails with clear content (e.g., invoice, meeting invite).
Use the manual trigger to process them.
Verify labels are applied correctly.

Activate for Auto Mode:
Toggle the workflow to Active.
New unread emails will be processed automatically every minute.

Notes

Dual Execution Modes**:
Auto Mode - Gmail Trigger polls inbox every minute for unread emails (real-time processing).
Manual Mode - Use the manual trigger to bulk process existing emails (adjust limit in "Get many messages" node).

AI Learning from Patterns**:
The workflow applies an 80% majority rule - if 80% or more of a sender's historical emails share the same label, the AI strongly prefers that category for new emails from that sender.
This creates intelligent sender-based routing over time.

Skip Already Labeled Emails**:
The "Check Label Existence" node prevents re-processing emails that already have an AI-assigned label.
Ensures efficient execution and avoids duplicate work.

Structured AI Output**:
Uses a Structured Output Parser to ensure the AI always returns valid JSON: { "label": "Category" }.
If uncertain, returns { "label": "None" } and the email stays in inbox.

Background Archiving**:
After labeling, emails are automatically removed from INBOX (archived).
Maintains a zero-inbox workflow while preserving emails under their labels.

Rate Limits**:
Google Gemini free tier: 15 requests/minute.
Adjust polling frequency if hitting limits.

Example Behavior

Minute 1**: New invoice email arrives → AI fetches sender history → 85% were labeled "Expenses" → applies "Expenses" label → archives email.
Minute 2**: Meeting invite arrives → No sender history → AI analyzes content (Zoom link, time) → applies "Meetings" label → archives email.
Minute 3**: Promotional email arrives → AI compares with "Trash Likely" examples → applies label → archives email.
Minute 4**: Already-labeled email detected → skipped silently.

Label Categories

| Label | Description |
|-------|-------------|
| Meetings | Calendar invites, Zoom/Meet links, appointments, scheduled events |
| Expenses | Bills, invoices, receipts, payment reminders, subscription renewals |
| Income | Payments received, payouts, deposits, earnings notifications |
| Notify / Verify | Verification codes, login alerts, 2FA codes, account notifications |
| Orders / Deliveries | Order confirmations, shipping updates, tracking numbers, deliveries |
| Inquiries | Business outreach, sales proposals, partnerships, cold emails |
| Trash Likely | Spam, newsletters, promotions, marketing blasts, ads |

> If no category fits clearly, the email returns "None" and remains in the inbox.

Customization

Change Polling Frequency**: Edit the "Gmail Trigger" node → pollTimes → mode (e.g., every5Minutes).
Adjust Email Limit**: Modify limit: 10 in "Get many messages" node for manual bulk processing.
Add Custom Labels**: Create in Gmail → Get ID → Update both Code nodes + AI system prompt.
Modify 80% Rule**: Edit the AI agent's system message to adjust the majority threshold.
Increase Tool Limits**: Change limit: 10 in Gmail tool nodes to fetch more historical data.

Author: Muhammad Anas Farooq