Inbox Triage Prompts: Overwhelmed vs Prioritized

A sales director at a commercial real estate firm returns from a two-day site tour to 214 unread emails: lease amendments, a tenant escalation, two broker commission queries, a city zoning notice, a board deck request, and roughly ninety newsletters. She pastes the whole inbox into ChatGPT, types “summarize my emails and tell me what to do first,” and gets back a friendly paragraph that mentions everything and prioritizes nothing.

Email triage is a sorting problem, not a summarizing problem. A summary keeps every message at the same weight; triage sorts them into what needs a decision now, what can wait, and what should never have reached the inbox at all.

Why a Stuffed Inbox Defeats a Loose Prompt

When you hand a model 214 messages and ask for a summary, the model has no reason to treat the zoning notice differently from a vendor newsletter. It compresses everything into prose, and prose is not a queue. The director still has to open the same emails to find the three that carry a deadline.

The signals that matter are buried in metadata and phrasing: a subject line with “URGENT” or “signature required,” a sender who is a client versus a list, a body that contains a date, a question, or a request for a decision. A triage prompt is built to extract exactly those signals and ignore the rest.

A Catch-All Prompt vs a Triage Prompt

Both prompts see the same inbox. One asks for meaning; the other asks for a work queue with buckets and owners.

Here are my 214 unread emails. Summarize them and tell me what I
should do first. [[inbox pasted]]

Role: executive assistant running inbox triage for a commercial real
estate sales director. Sort, do not summarize. Do not draft replies.

Buckets (one per email):
- REPLY TODAY: needs my decision, approval, or a yes/no.
- DELEGATE: someone on my team should own it; name the role.
- WAITING: I am blocked on someone else; note the follow-up date.
- FYI: relevant, no action.
- ARCHIVE: newsletters, digests, cold outreach.

Rules
- Output one row per email: sender, trimmed subject, bucket, one-line
  why, suggested next action, deadline if stated. No prose paragraphs.
- Sort so REPLY TODAY comes first, then DELEGATE, WAITING, FYI,
  ARCHIVE.
- Tag any email about leases, commission, legal, zoning, or the board
  with ESCALATE in the why column.
- After the table, list the five emails to handle in the next 30
  minutes and the single reply that unblocks the most people.
- If a deadline is implied but not stated, write "deadline not stated"
  instead of guessing.

Inbox:
[[emails pasted as sender, subject, date, body]]

The first prompt produces an essay. The second produces a sortable table plus a short action list. The director reads five rows and starts working; the other 209 are already filed.

How Triage Restores Control of the Day

• Buckets replace judgment with rules. “Needs my decision” is a test, so each email lands somewhere instead of floating as “important.”
• Delegation is explicit. Naming a role, not a person, turns a message into an action someone else can take without a hallway conversation.
• Waiting items get dates. Flagging a follow-up date prevents the quiet failure where a broker waits two weeks for a reply.
• Escalation is keyword-triggered. Lease, zoning, and commission emails are exactly the ones a sales director cannot let sit, so the prompt forces them up.
• The five-item list creates a starting point. A ranked next-30-minutes list removes the paralysis that a 214-message pile creates.
• No drafting means no false confidence. Keeping the model to sorting stops it from sending a reply the director would never approve.

For the director’s inbox, the output might place a tenant escalation and a lease amendment under REPLY TODAY, a commission query under DELEGATE with an ESCALATE tag, and the city zoning notice under FYI with a review date. Ninety newsletters collapse into a single ARCHIVE line.

Buckets That Match Your Role

The exact buckets change by role. An engineer might need “Bug,” “Deploy,” and “Question.” A recruiter might need “Candidate reply” and “Hiring manager.” The principle holds: three to five buckets, each defined by a test the model can apply, with exactly one bucket for genuine no-action mail.

Adapt the Triage Prompt to Your Own Inbox

Role: assistant doing inbox triage for a {{role}}. Sort, do not summarize.
Do not draft replies.

Buckets:
- {{BUCKET 1: needs my decision}}
- {{BUCKET 2: delegate, name a role}}
- {{BUCKET 3: waiting on someone, note a date}}
- {{BUCKET 4: FYI, no action}}
- {{BUCKET 5: archive}}

Rules
- One row per email: sender, trimmed subject, bucket, why, next action,
  deadline.
- Sort by bucket, most urgent first.
- Tag {{keywords that matter to you}} with ESCALATE.
- After the table, list the {{n}} emails to handle in the next 30 minutes.
- Write "deadline not stated" rather than guessing.

Inbox:
{{emails as sender, subject, date, body}}

Two habits make this reliable. First, paste sender, subject, date, and body; without dates the model cannot honor the deadline rule. Second, keep the bucket names stable across weeks so your own review stays comparable. If you already run a weekly planning ritual, the triage table feeds straight into it, and pairing it with a fixed calendar routine keeps the follow-ups from falling through.

Sources & Notes

The triage approach reflects common inbox-management practice and observed prompt behavior with GPT-4-class models in 2024–2025. The sales director and the 214-message inbox are illustrative. Sorting improves focus but does not replace judgment on legal or financial matters, which is why the prompt flags those for escalation. Internal references: Inbox and Calendar Prompts, Productivity Prompts List, Weekly Work Plan ChatGPT Prompt.