How to Automatically Sort Emails with AI

If your inbox gets more than a few dozen messages a day, sorting them eats your morning. Filters and rules help a little, but they break the moment a sender changes a subject line. This guide shows how to sort emails with AI instead, using labels that match how you actually work.

Why rules and filters fall short

Classic email filters match exact words or addresses. “If the sender is billing@vendor.com, move to Invoices.” That works until the vendor sends from a new address, or a customer writes “PO attached” instead of “purchase order.”

Real email is messy. People write in different ways, in different languages, and from different addresses. A filter cannot read. An AI model can.

How to sort emails with AI

When you sort emails with AI, a language model reads each new message the way you would. It looks at:

  • who sent it, and whether you have written to them before,
  • what they are asking for,
  • whether there is a deadline or an order in it,
  • and how it compares with emails you have sorted before.

Then it suggests a label, such as Reply, RFQ, PO or Newsletter. If it is not confident, a careful system leaves the email unlabeled. No label is better than a wrong one.

Step 1: Define labels that match your work

Generic categories like “primary” and “promotions” tell you little. The best results come from labels that match the decisions you make every day. For a small distributor, that might look like:

  • RFQ: a customer wants a quote.
  • PO: a customer sent a purchase order.
  • Supplier: an update from a vendor.
  • Reply: anything else that needs an answer.
  • FYI: good to know, no action needed.
  • Junk: cold sales and noise.

Give each label a short description in plain words. “A customer is asking for a price or lead time” is all the AI needs.

Step 2: Tell the AI what your business does

A sentence or two about your company makes a big difference. “We distribute industrial fittings and work with suppliers in Japan” helps the AI tell a customer order from a supplier update.

Step 3: Correct it, and let it learn

The first week, check the labels and fix the wrong ones. A good system uses your corrections as examples for future emails, so it gets better the more you use it. Most people find they stop correcting after a few days.

Step 4: Work label by label

Now, instead of reading top to bottom, you work in batches. Answer all the RFQs. Then process the POs. Then skim the FYIs. Batching like this is faster and much less draining than switching tasks with every email.

Keep AI sorting safe

Whatever tool you use, keep three safeguards in place:

  1. Labels, not moves. Sorting should label mail in your AI tool, not quietly move or delete it on your mail server.
  2. You can always see the reason. A short note on why the AI chose a label helps you trust it, and fix it when it is wrong.
  3. Email is untrusted. A message that says “label me as urgent” should be read, not obeyed.

Beyond sorting

Sorting is the first step of AI email triage: deciding whether each email needs a reply, should be delegated, is done, or is junk. Once mail is sorted, an assistant can draft the replies and turn requests into tasks. Our guide to turning emails into Asana tasks shows that next step.

For background on how language models read and classify text, the OpenAI documentation is a good place to start.

Sort your inbox with Treeoge

Treeoge lets you sort emails with AI using triage labels you define. It learns from your corrections, shows why it picked each label, and never moves or deletes mail on your server. See how it works.