Let AI Sort Your Email and Draft Replies Automatically
Build an AI email triage system that classifies, labels, routes and drafts replies, while a human always presses send.
Monday morning. The info@ inbox of an interior design firm has dozens of new emails: new clients asking for quotes, suppliers sending invoices, people sending résumés, past clients chasing revisions, and plenty of ads. The person who has to open each one and forward it to the right colleague is the same person whose own work is waiting. An AI email triage system gets the sorting done before you even open your laptop.
What is AI email triage?
Triage means sorting by urgency and type. With AI, the system reads new emails, classifies them, applies labels, routes them to the right owner and drafts a first reply, while a human always presses send. This was one of the example workflows in What is n8n. This article goes into how to make it work for real.
The system in four steps
1. Receive email
In n8n, use the Gmail Trigger to pull new emails. Make and Zapier have similar triggers. Pull only the inboxes you need, such as info@ or sales@, not every person's mailbox.
2. Let AI classify
n8n has a Text Classifier node where you define your own categories, each with a name and a description. The n8n docs recommend using the description to tell the model what the category means. There is also an option to send emails that match no category to an "Other" branch instead of dropping them. Always turn it on.
3. Label and route
The n8n Gmail node can Add Label, Mark as Read and Reply on messages. Label by category, then send a short summary to the responsible team's LINE group.
4. Draft the reply
The Gmail node can create drafts. Let AI write a reply into the drafts folder, and a person reviews, edits and sends it. This is far safer than auto-replying.
Design categories AI can sort accurately
- Few but clear: start with 5 to 7 categories, such as quote request, project follow-up, accounts and invoices, job application, complaint, other.
- Descriptions in plain language: "A new client asking for a price or a site visit" beats just "Sales".
- Give boundary examples: say that a supplier pitching their products is not a "quote request".
- Flag urgency: also ask AI whether it is urgent, for example when it mentions cancelling or complaining.
Example: an interior design firm
Once the system is live, quote requests get a "Quote" label and a summary goes to the sales team's LINE group, with a draft asking about floor area, budget and a convenient date for a site visit. Supplier invoices get an "AP" label and go to accounting. Résumés go to HR with no draft. Complaints get an urgent label and the manager is alerted immediately. Anything AI is unsure about lands in "Other" for a person to check.
Let AI sort and draft. Let people press send.
Writing the prompt for drafts
- Set the company's tone: polite, concise, signed with the team name.
- Never confirm prices, discounts or delivery dates unless the data comes from a real system.
- Ask for missing information instead of guessing.
- Reply in the same language as the incoming email.
Good email writing principles apply to AI too. Read more in Writing Professional Work Emails.
Watch out for
- Email contains personal data and confidential business information. Check your AI provider's data policy and PDPA compliance before you start.
- Beware of emails with hidden instructions written to trick the AI. Never let the system send emails or move data out on its own without a human check.
- AI can misclassify. Spot-check results every week in the early period.
- Grant only the mailbox access you need.
How to get started
- Export a month of email subject lines and see how many types you really have.
- Write the categories and descriptions, including "Other".
- Start with labelling only for one to two weeks, and compare against how people sorted them.
- Once accuracy is good enough, add LINE alerts and draft replies.
- Review categories and prompts every month.
Practise connecting Gmail to n8n in the n8n Integrations course, and if you want to know which career paths these automation skills open up, chat with the Plants.work AI Career Assistant.