Email Triage: How to Triage an Inbox That Never Empties
Put every mailbox on one board and let AI categorize and de-spam each message.
Email triage is the practice of sorting incoming mail by what it requires from you, before you start replying to any of it. The word is borrowed from emergency medicine, where triage means deciding the order of treatment rather than treating everyone at once. Applied to an inbox it means the same thing: one fast pass that classifies every message, then a separate pass where you actually work. Doing those two jobs in one pass is why most inboxes never empty. Here is a triage system that holds up when the mail arrives at four addresses instead of one.
What does email triage mean?
Triage means deciding what each message is and what it needs, and nothing else. You are not writing replies during triage. You are labeling: this needs a real answer, this needs two minutes, this is a receipt to file, this is a newsletter, this is noise. The output of triage is a sorted pile, not an empty inbox. The reason to separate it from replying is that the two use different kinds of attention. Classifying is fast and mechanical. Replying is slow and requires context. Interleaving them means you pay the cost of switching between the two hundreds of times a day.
Most productivity advice skips this distinction, which is why it fails on a heavy inbox. "Touch each email once" is good advice for a light inbox and impossible on a heavy one, because a message that needs 40 minutes of work cannot be handled the moment you first see it. Triage is the honest version: touch each email once to categorize it, then handle categories in blocks.
How do you triage an inbox?
Four categories are enough. More than that and you spend the triage pass deciding which bucket something belongs in, which defeats the purpose.
| Category | Test | What happens to it |
|---|---|---|
| Reply now | Under two minutes and you have everything you need | Answer it during triage, then it is gone |
| Reply later | Needs real thought, information you do not have, or a decision | Goes to a list or a folder with a time blocked to do it |
| Reference | No action, but you will want it again: receipts, confirmations, contracts | Archived, findable by search, never read twice |
| Noise | Newsletters, notifications, marketing, anything you would not miss | Archived or unsubscribed, ideally never in the inbox to begin with |
Run the pass top to bottom without skipping ahead to the interesting messages. Skipping is what turns a ten-minute triage into an hour, because the interesting message pulls you into replying and the pass never finishes. Set a timer if that helps. The goal is a categorized inbox, and you should be able to triage a hundred messages in well under fifteen minutes once the categories are automatic.
How often should you triage your email?
Two or three fixed times a day works for almost everyone: once mid-morning after you have done something real, once after lunch, once before you stop. Continuous triage, meaning you look every time something arrives, is the failure mode. It converts a fifteen-minute job into a background process that runs all day and never finishes, and it makes deep work impossible because you are never more than a few minutes from an interruption.
The exception is genuinely time-sensitive mail, and the fix there is not to check constantly but to make the urgent stuff announce itself. Notifications for a handful of specific senders, or a rule that flags anything from your largest client, gets you the responsiveness without the constant checking. Everything else can wait until the next pass, and almost all of it can.
What is an email triage system?
A system is just triage that survives you being busy. Three things make it durable. First, the categories are fixed, so no decision is required about where something goes. Second, the noise is dealt with structurally rather than by hand: unsubscribed, filtered, or routed away, so it never reaches the triage pass at all. Third, the sorting is not something you have to remember to do, because it happens as mail arrives rather than when you get round to it.
That third point is where systems usually break. A ruleset you maintain by hand decays: every new sender, newsletter, and vendor needs a new rule, and after six months you are maintaining filters instead of reading mail. This is the argument for having the classification done automatically, and it is why AI email management has become a category. A model that reads each message and decides what it is does not need a rule written for every sender it has never seen.
Does triage still work with multiple email accounts?
Not without one change, and this is the part most advice ignores. Triage assumes a single queue. If you have four mailboxes, you do not have one queue, you have four, and you triage them at different times with different amounts of attention. The account you check least becomes the one where something important goes unnoticed for a week, which is the exact failure triage was supposed to prevent.
The fix is to merge the queues before you triage them, not after. Every mailbox on one board, sorted in one pass, with replies still leaving from the correct address. That is the practical case for combining email accounts into one inbox: not tidiness, but making triage possible at all when the mail arrives at more than one address. The alternative, checking four inboxes three times a day, is twelve triage passes, and nobody sustains that.
Can AI do email triage for you?
Partly, and it is worth being precise about which part. Classification is the half that automates well: deciding whether a message is a receipt, a newsletter, a notification, a real piece of work, or spam is a judgment a model makes reliably and instantly, across every message, without a rule per sender. That is the tedious half and the half that does not benefit from your attention.
Deciding what to do is the half that does not automate, because it depends on context a model does not have: which client matters this quarter, which thread is politically delicate, which request you are going to decline. So the useful division of labor is that the machine sorts and you decide. When you open a board where every message already carries a category and the spam is gone, the triage pass stops being classification work and becomes what it should have been: a short scan of things that actually need you.
This is what Inboxes does. Connect every mailbox over IMAP, and an AI model reads each message as it arrives and assigns it a category, work, personal, receipts, newsletters, or notifications, plus a spam verdict, across every account at once. There is a related problem it does not solve, which is finding the one message you half-remember from eight months ago. That is a search problem rather than a triage problem, and it is the kind of thing search that reaches across everything a company writes down is built for.
What triage does not fix
Triage is a sorting discipline, not a volume control. If you get 300 real messages a day that each need a considered answer, no system fixes that; the problem is the volume and the answer is delegation, a shared address, or fewer commitments. Triage also will not save an inbox you use as a to-do list, because a to-do list sorted by arrival time is not a to-do list. And it does nothing about a backlog: 20,000 unread messages are not a triage problem, they are an archaeology problem, and the honest move is to archive the lot and start clean from today.
What triage does fix is the daily churn, which for most people is the actual complaint. Fifteen minutes of sorting, two or three times a day, against a merged and pre-categorized queue, is a sustainable relationship with email. If you want the longer version of the destination rather than the method, start with what a unified inbox is and how it differs from just stacking accounts on top of each other.