Step 2: Connect Google Sheets

Configure the two Sheets nodes to log classified emails and classification errors.

1. Open Record All Emails and select the target spreadsheet in Document, then set Sheet Name to Sheet1.
2. Verify the column mappings include values like {{ $json.subject }}, {{ $json.category }}, and {{ $json.actionItemsJson }}.
3. Credential Required: Connect your googleSheetsOAuth2Api credentials to Record All Emails.
4. Open Record Classification Errors and set Sheet Name to Error Log.
5. Credential Required: Connect your googleSheetsOAuth2Api credentials to Record Classification Errors.

Step 3: Set Up AI Classification and Parsing

Configure the AI agent, language model, and structured parser that classify and extract entities from emails.

1. Open Categorize and Pull Entities and set Text to Sender: {{ $json.senderName }} ({{ $json.senderEmail }})\nSubject: {{ $json.subject }}\nBody: {{ $json.bodyText }}.
2. In OpenAI Conversation Model, set Model to gpt-4.1-mini, Max Tokens to 1500, and Temperature to 0.2.
3. Open Structured Result Parser and confirm the JSON Schema Example matches your desired fields (category, priority, sentiment, entities, actionItems, summary).
4. Credential Required: Connect your OpenAI credentials to OpenAI Conversation Model.
5. Note: Structured Result Parser is an AI tool sub-node; credentials should be added to OpenAI Conversation Model, not the parser.

Step 4: Configure Processing and Validation Logic

Clean the email payload, validate AI results, and route valid classifications.

1. Open Sanitize Email Payload and keep the JavaScript logic that extracts messageId, senderEmail, subject, and a bodyText limited to 3000 characters.
2. Open Validate AI Output and keep Mode as runOnceForEachItem to enforce valid categories, priorities, and sentiments.
3. Open Verify Classification Result and ensure the condition checks {{ $json.classificationValid }} is true.
4. Open Route by Email Category and confirm the rules compare {{ $json.category }} to values like Inquiry, Support Request, and Urgent.

Validate AI Output outputs to both Verify Classification Result and Flag High Priority in parallel.

Step 5: Configure Output and Notification Actions

Apply Gmail labels, record results, and send urgent alerts to Slack.

1. For each Gmail label node (Apply Label: Inquiry, Apply Label: Support, Apply Label: Newsletter, Apply Label: Marketing, Apply Label: Personal, Apply Label: Urgent, Apply Label: Spam, Apply Label: Invoice, Apply Label: Meeting), set Operation to addLabels and Message ID to {{ $('Sanitize Email Payload').first().json.messageId }}.
2. Credential Required: Connect your Gmail credentials (Google OAuth) to all Gmail label nodes.
3. Open Flag High Priority and ensure conditions include {{ $json.priority }} = High, {{ $json.category }} = Urgent, or {{ $json.sentiment }} = Urgent.
4. Open Send Urgent Slack Alert and keep Text as the multi-line template starting with 🚨 *URGENT EMAIL ALERT*.
5. Credential Required: Connect your Slack credentials to Send Urgent Slack Alert and select the target Channel.

Each label node routes into Record All Emails, ensuring all classified emails are logged after labeling.

Step 6: Add Error Handling

Invalid AI outputs are captured and recorded for review.

1. In Verify Classification Result, ensure the false output is connected to Record Classification Errors.
2. Confirm Record Classification Errors maps fields like {{ $json.classificationError }} and {{ $json.bodyText.substring(0, 500) }} for quick troubleshooting.

Step 7: Test and Activate Your Workflow

Run a controlled test to verify labeling, logging, and alerts before turning it on.

1. Manually execute Gmail Poll Trigger with a sample unread email in your inbox.
2. Check that labels are applied in Gmail and that a new row appears in Record All Emails.
3. If you send a high-priority email, confirm Send Urgent Slack Alert posts the formatted message.
4. For invalid or malformed AI results, verify that Record Classification Errors logs the issue.
5. Once validated, toggle the workflow to Active for continuous automation.