New Gmail emails sorted by AI and logged to Google Sheets

New Gmail emails sorted by AI and logged to Google Sheets

An n8n workflow that sorts your new unread email and keeps a log. Every minute it picks up new unread messages in
Gmail. A model labels each one as urgent, important, informational or spam and writes a one-sentence summary. The result
goes into a Google Sheet. It is the safe first step: nothing is moved, deleted, labelled or answered.

What is inside

• workflow.json: the workflow as n8n exports it, six nodes. One is a sticky note. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The AI step was an n8n-nodes-base.openAi node at version 1.8. That node type only exists at versions 1 and 1.1, and
  its options differ. AI: Classify is now n8n's OpenAI node, which has version 1.8, with its Message a Model operation and
  JSON output turned on.
• The model was gpt-4o-mini, which OpenAI shuts down on 14 January 2027. It now asks for gpt-6-luna, the replacement
  OpenAI names. The temperature of 0.2 and the limit of 200 tokens are gone. The model's page does not list a temperature
  setting, and OpenAI counts a reasoning model's thinking against the output limit, so a small cap can cut the answer off.
• Parse Email handled only the first email of each check. When several emails arrive between two checks, the others were
  lost. It now handles all of them, and Merge Result pairs each answer with its email.
• The snippet was read from a field that the full email output does not have, so the model saw only the sender and the
  subject. It now falls back to the start of the plain text.
• Merge Result expected the answer as text. The OpenAI node now returns it as an object, so the code takes either.
• The sheet id came from an environment variable, which only a self-hosted n8n can set. It is now
  the placeholder <YOUR_SHEET_ID>. The tab is Email log, and the node lists its five columns.
• The sticky note has a shorter title and a plainer last line.

Before you start

• A Gmail account you can connect to n8n with OAuth2.
• An OpenAI API key.
• A Google Sheet with a tab called Email log. Row 1 holds these names: Date, From, Subject, Category and
  Summary. If one is missing, the sheet step stops.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Gmail OAuth2 credential and select it in Gmail: New Email.
3. Add an OpenAI credential and select it in AI: Classify.
4. Add a Google Sheets OAuth2 credential and select it in Log to Sheets. Replace <YOUR_SHEET_ID> with your sheet's id,
   the part of its address between /d/ and /edit.
5. Publish the workflow, then send yourself a test email from another address.

How a run goes

1. Gmail: New Email fetches new unread messages and parses them in full.
2. Parse Email keeps the message id, the sender's address and name, the subject, the first 500 characters of the body and
   the date.
3. AI: Classify asks the model for JSON with a category and a summary. The category is urgent, important,
   informational or spam.
4. Merge Result joins each answer with its email. If an answer cannot be read, the email is logged as informational with
   no summary.
5. Log to Sheets appends a row to the Email log tab.

Change the result

• Which mail is looked at: the Read Status and other filters in Gmail: New Email.
• The categories and the summary: the system message in AI: Classify, and the fallback in Merge Result.
• How much of the body the model sees: the substring length in Parse Email.
• The columns: the mapping in Log to Sheets and row 1 of your tab.

Good to know

• The workflow marks nothing as read. The unread filter only limits which mail is fetched at each check.
• Each new email is one call to OpenAI. It sends the sender, the subject and the start of the body.
• The model's category is a guess. Read the first rows before you trust them.
• Mindket did not run this workflow against Gmail, OpenAI or Google Sheets. These notes follow n8n's published node code
  and OpenAI's pages, read on 12 November 2026, and the two Code steps were run on sample emails on a local computer.