Sorting and Labelling Emails With AI

Start with the labels you want, then decide who or what applies them. A short list is plenty: sales, support, billing, suppliers, junk and "needs a person".

Labels are not folders

Google points out that labels are different from folders.¹ That matters because a label can sit on a message without moving it anywhere. Gmail also allows plenty of room: up to 5,000 labels.¹

Keep your list short anyway. A long one gives the AI more ways to be wrong. It also gives your team more to remember, and people forget. Fewer labels, used well, beat many labels used badly.

Three ways to sort

Google's built-in tabs are a good example of the middle option. In the default inbox type, Gmail sorts mail into tabs on its own: Primary, Social, Promotions, Updates and Forums.² You can only pick from those. So the tabs won't give you a "billing" or "sales" label of your own.²

If a tab gets a message wrong, drag it to the right one. Google says that over time this helps Gmail learn your preferences.²

Where AI comes in

Google says you can ask Gemini to add labels or delete several emails at once.³ Gemini acts when you ask it. Google says it shows a confirmation card first, and nothing is labelled until you confirm. It can only add labels that already exist in your account, and it can't create new ones.³ The feature needs an eligible Google Workspace or Google AI plan.³ Google's page doesn't say Gemini sorts new mail into your labels on its own. If you use other tools, read the maker's own page to see what they can label and which plan you need.

Wikipedia notes that some advanced filters use statistical techniques to organise mail, rather than fixed rules.⁴ That's the basic idea behind AI sorting: it guesses from patterns, so it can be wrong.

How to set it up

• Write down your labels and, for each one, a one-line description in plain words. "Billing: questions about an invoice, a payment or a refund."
• Let plain rules take the obvious mail first, such as your accounting software's notices.
• For the mail that's left, ask the AI to suggest a label, using your one-line descriptions, and check what it proposes before you confirm.
• For the first few weeks, have a person check a sample of each label.

Treat "junk" with care. Don't let any tool delete mail on its own at the start. Not yet. Send it to a folder you can look through, so a real customer message doesn't vanish.

What to watch for

Mistakes cluster in a few places, and once you know where they are, a quick weekly look at those spots catches most of them before a customer notices.

• Messages that fit two labels. Decide which label wins, for example "complaint" over "sales".
• New kinds of mail, such as a supplier's change of bank details. Those deserve a human look whatever the label says.
• Labels the team has stopped using. Remove them.

Rules before AI are covered in simple inbox rules. To catch time-sensitive mail, see flagging urgent emails.