10 Prompts That Sort Your Inbox

TL;DR. A good category prompt says what belongs, what looks similar but doesn’t, and why. Below are 10 you can copy straight into AI Email Filter and adjust to your own inbox, each with a short note on why it’s written that way.

Related reading: AI-rewritten forwarding · Newsletter overload

AI Email Filter classifies each email (or each sender, if you’re using Sender Smart Categories) against categories you describe in plain language — no filter syntax. The categories below are illustrative starting points, not a fixed feature list; adapt the wording to match how your own inbox actually looks. What makes each one work is the same pattern every time: a clear positive definition, an explicit boundary against the nearest look-alike category, and — where it matters — the action that should follow.

1. Invoices and bills

“Invoices, bills, and payment reminders that require me to pay something. Exclude payment confirmations for purchases I already made and marketing emails about discounts.”

Why it works: “Invoices” alone is ambiguous — a receipt and an invoice look similar to a classifier if you don’t separate “money I owe” from “money I already spent.” The exclusion clause does the actual work here.

2. Cold outreach and sales pitches

“Cold outreach, LinkedIn-style connection pitches, or anyone trying to sell me a product or service I didn’t ask about. Exclude replies to threads I started and messages from people I’ve corresponded with before.”

Why it works: The hard part of catching sales pitches isn’t the pitch itself, it’s not catching a legitimate reply from someone you already talked to. The second sentence protects against exactly that false positive.

3. Receipts and order confirmations

“Order confirmations, shipping notifications, and payment receipts for things I’ve already bought. Exclude promotional emails from the same stores, even ones I order from regularly.”

Why it works: This is the classic same-sender problem — a shop sends both receipts and promotions from one address. Naming the store-promo overlap explicitly is what keeps the two apart.

4. Travel and bookings

“Flight confirmations, hotel bookings, itinerary changes, and check-in reminders for trips I’ve booked. Exclude travel deal newsletters and ads for destinations I haven’t booked.”

Why it works: Travel inboxes get flooded with “cheap flights to X” marketing that shares vocabulary with real booking confirmations. The exclusion draws the line on intent, not keywords.

5. Account and security alerts

“Security alerts, login notifications, password reset requests, and account warnings from services I actually use. Exclude generic ‘protect your account’ marketing from services trying to upsell a security product.”

Why it works: Real security alerts need visibility, fast — this category should usually stay in the inbox rather than get filed away, so keeping it narrow matters more than keeping it broad.

6. Newsletters worth reading

“Newsletters and digests from sources I’ve deliberately subscribed to and actually want to read. Exclude one-off promotional blasts and re-engagement emails from lists I signed up for once and forgot about.”

Why it works: Not all newsletters deserve the same treatment. This prompt keeps the ones you value distinct from the ones cluttering the same folder, which matters if you’re trying to actually read a curated list rather than archive everything with “newsletter” in the subject.

7. Personal messages

“Messages from friends or family, written personally rather than sent through an automated system. Exclude automated event invites, group notifications, and anything sent to a mailing list.”

Why it works: “Personal” as a category tends to collect automated noise (birthday reminders, group-chat digests) unless you explicitly separate a human writing to you from a system writing on someone’s behalf.

8. A specific project

“Anything related to the [Project Name] launch — updates, questions, deadlines, or decisions — regardless of who sends it or which team they’re on.”

Why it works: This is the prompt style that a sender-based rule fundamentally cannot express: the category is defined by topic, not by an address, so a new stakeholder joining the project gets sorted correctly on their first email with no rule to update.

9. Customer support requests

“Direct questions, complaints, or support requests from customers about something not working. Exclude internal team discussions about support tickets and automated ticketing-system notifications.”

Why it works: A support inbox mixes real customer voices with the tooling built around them (ticket-system CCs, status updates). Separating the two keeps the category focused on messages that actually need a human reply.

10. Spam-ish and low-value bulk mail

“Unsolicited pitches, scams, generic ‘exclusive opportunity’ emails, and bulk mail with no clear connection to me or my work. Exclude anything from a sender I’ve replied to before, even if the subject line looks promotional.”

Why it works: This category is meant to catch what a provider’s spam filter lets through — genuinely low-value bulk mail — without catching legitimate mail that happens to share vocabulary with spam. The reply-history exclusion is the safety valve.

Turning a prompt into an action

A category description says what a message is; it’s a separate step to decide what happens to it. For newsletters you’ve deliberately kept but don’t want cluttering your inbox, forwarding with an AI-generated summary instead of the raw mail is often more useful than just filing it — see AI-rewritten forwarding for how that works, and newsletter overload for a fuller setup built around exactly this problem.

Before you trust any of these on autopilot

Run a dry test against real mail before enabling automatic actions on a new category, especially ones like #5 or #9 where a false negative has real cost. If a category is consistently wrong on one type of message, the fix is almost always a sharper exclusion clause, not a longer positive description — every prompt above follows that shape for a reason.