Triage your shared inbox with AI

Shared inboxes fail in a predictable way: everyone can see the email, so nobody owns it. This recipe gives an AI assistant three jobs. Classify every email into a short list of categories you define. Route it to a named owner using rules you write once. Draft a reply, but only for categories you have marked safe, and never with invented facts — the prompt forces a [BLANK] wherever the AI lacks a real order number, date, or price.

It is built for a business inbox getting enough daily volume to hurt, run by a team with no AI tooling beyond a chat assistant. If that is you, this works on day one.

The recipe is five steps. Sort your last 30 emails by hand to find your real categories, and merge any two that route to the same person. Fill in the routing rules template, including which categories are safe to draft. Paste the triage prompt into your assistant, ideally as saved instructions so you only paste emails after that. Then test: run 20-30 already-handled emails through and compare the AI's calls to what your team actually did, tightening category definitions until you would act on the output without wincing.

Once it passes, make it a routine. Batch new emails through once or twice a day. A human reads every draft before it sends, and nothing is ever auto-sent.

Honest limit: this recipe has a person in the middle, pasting emails in and drafts out. That is the right training wheel, and for lower-volume inboxes it may be all you ever need. It stops being enough when the pasting becomes its own chore, or when you want every inbound email processed automatically, wired into your helpdesk or CRM, with logging and escalation you can audit. That is a build, not a prompt. When you hit that wall, look at what we install or book a free AI opportunity audit and we will tell you whether you are actually there yet.