← All use cases

中文

Personal productivity · Claude

Let Claude sort and archive a test email inbox every morning

Idea + starter prompt, not run by us yet

For: People buried in email who want a trial run before trusting AI

A PCWorld writer tried this on a spare Gmail account, not his work mail. A morning automation in Claude's Cowork tab checks the last 24 hours, tags each thread Important or Archiveable, then archives that second group and tags receipts. A triage summary comes with it. Reply drafts are for business and school contacts only.

Steps

  1. Use a spare or test Gmail account, not your work inbox, and connect it to Claude.
  2. Write the rules in plain words: what counts as Important, what is Archiveable and which senders get draft replies.
  3. Ask for a plan and a dry run that only lists what it would label or archive.
  4. After a day, check the archive and the labels. The author's first day filed receipts and shrank the inbox but produced no reply drafts yet.
  5. Know the limits. The Gmail link can draft but not send without approval. It can move mail to trash but not erase it for good, and trash keeps it 30 days.

Starter prompt

Starter prompt. Change the [brackets] first.

Act as my email triage helper on my test Gmail account [test account]. Look only at the last 24 hours of mail. First do a dry run. List each thread and say whether you would label it Important, Archiveable or Receipt, with a one line reason. Don't archive, label, trash or send anything until I say go, and never delete mail. After I approve, apply the labels and archive the Archiveable threads. Then write a short triage summary and draft, but do not send, replies for [business and school contacts]. Leave mail from friends and family alone. Don't ask for my password. Use only sample messages with invented names in this test account. Do not copy message bodies or contact details into other files or services.

Original post: pcworld.com · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.