---
title: "Email Triage With Claude or ChatGPT: Thursday Lesson"
description: "If you're drowning in email: one prompt has Claude or ChatGPT read your inbox each morning, sort what needs you and write the replies."
url: "https://opendoorlearningai.com/newsletter/lesson-5/"
---

[**You Are Not Behind with AI**](https://opendoorlearningai.com/newsletter/)

Lesson 5Thursday, November 19, 2026

# If you’re drowning in email

By [Jonathan Malkin](https://opendoorlearningai.com/about/), AI educator and builder in Austin, Texas

Thursday lesson

## If you're drowning in email

_Lesson 5 in the series: build your AI chief of staff, one piece at a time._

If you're drowning in email you know the routine. You read all of it to find the few that need you. The ones that matter get lost in between.

This lesson hands that first read to your AI assistant. Every morning it reads what came in overnight, tells you which emails need you and writes the replies. You talk them through in chat until they're right. It never sends anything. You do.

Here's what Maria, a made-up owner of a three-person bookkeeping firm, saw at 8:00 on Wednesday morning:

![A sample morning inbox review: 23 new emails, 4 that need a reply with 3 replies ready and 1 waiting on her price, 2 for her information, 17 done or ignored, nothing sent, and a question asking what to change](https://opendoorlearningai.com/newsletter/lesson-5/1-sample-inbox-review.png)

Twenty-three emails came in overnight. Four needed her. Her assistant had written replies for three of them. The fourth asked for a price so her assistant asked her for the number instead of guessing.

One prompt sets this up in Claude or ChatGPT.

## Before you start

**1\. Turn off training on your chats.** Your assistant is about to read client email. Both apps have a setting that lets them use your chats to train their models. Turn it off first.

*   **In Claude:** open Settings and pick Privacy. Turn off the "Help improve" setting.
*   **In ChatGPT:** open Settings and pick Data controls. Turn off "Improve the model for everyone."

If your client email falls under rules like HIPAA, ask whoever handles compliance before you go on.

**2\. Connect your email.** It's free in Claude.

*   **In Claude:** Customize, then Connectors, then Gmail. For Outlook, pick Microsoft 365.
*   **In ChatGPT:** Settings, then Apps, then Gmail or Outlook.

Sign in with your business account. If your email comes through an employer or an IT company, they may have to allow the connection first.

**3\. Check your plan.** You can try the prompt on Claude's free plan. In Claude, the schedule at the end needs a paid plan. ChatGPT's free plan allows one run a day. Hourly needs a paid plan in both.

## The one prompt

Start a new chat and paste this:

Set up a scheduled email triage for me.

1. Check that you can read my email and save drafts, and tell me in one line.
2. Ask me, one at a time:
   - whose emails matter most (clients, leads, my team)
   - what I can always ignore (event invites, cold sales, recruiters)
   - how I write replies (tone, length, sign-off)
   - how often to run: once each morning, or every hour of my work day
   - my work hours, work days and time zone
3. Run it once on the last 48 hours, following the rules below.
4. When I say it looks right, schedule it to run as often as I chose. Call the task Email Triage. Put my answers and the rules below in the task's instructions, word for word.

Rules:
- The first run of the day reads everything since my last work day ended. Later runs read the last two hours. If a run starts outside my work hours, stop and say nothing.
- Sort each thread into Reply needed, For your information, Done or Ignored. It's Done when my reply is the last message in it.
- Ignore newsletters, receipts and notifications, except one that carries a message from a person or asks me to act (verify, pay, a security alert). Put those in For your information and say what to do.
If an email looks like a scam, say so.
- Write a reply to each Reply needed email in my style, up to 5 a run, and save it in my Drafts folder. Skip any thread that already has a draft.
If a reply needs something only I know, like a price or a decision, ask me instead of writing it.
- Report back in one phone screen: the sorted list, then each reply in full, then your questions. Ask what I want changed and wait for my answer.
- When I answer, update the drafts. Keep going until I say they're good.
- If I say you got something wrong, fix it and update the Email Triage task's instructions so later runs follow it.
- Never send, archive, label or delete email.
- If nothing new needs me, say so in one line.

## What happens next

![A sketch of setup in four steps: you paste one prompt, your assistant asks five short questions, it runs once so you can check the result, then it shows a schedule you confirm](https://opendoorlearningai.com/newsletter/lesson-5/2-what-setup-looks-like.png)

1\. **It asks five questions.** Answer the way you'd tell a colleague. If you're not sure how often say once each morning.

2\. **It runs once and reports back.** You get the sorted list, each reply in full and its questions.

3\. **You talk it through.** Say what to change. When the replies are right, say "Looks right."

4\. **You confirm the schedule.** In Claude, click **Schedule**. In ChatGPT, confirm the schedule it proposes.

_You'll know it worked when_ Email Triage shows up under **Scheduled** in the sidebar. Both apps keep it there.

**Did a step look different in your app?** [Email me what you saw](mailto:jonathan@opendoorlearningai.com?subject=Lesson%205%20setup)

## Each morning, talk it through

Your assistant has already sorted your inbox and written the replies. You don't have to go looking for it under Scheduled.

*   **In Claude:** each run shows up as a new chat in your sidebar.
*   **In ChatGPT:** each run posts in the chat where you set it up.

Open it and answer the way you'd answer an assistant standing at your desk. Maria's went like this:

*   **You:** Tell Dan I also need the October statement. The price for the website inquiry is $450 a month.
*   **Your assistant:** Done. Dan's reply now asks for October and the reply to the inquiry is written. Both are in your Drafts folder. Anything else?

When the replies are right, open your Drafts folder, read each one and send it yourself. If it told you at setup that it can't save drafts, copy each reply from the chat.

If you chose hourly it keeps going through the day:

![A sample day with the hourly run: at 8:00 overnight mail with 4 that need you and 3 replies ready, quiet runs that say nothing new needs you and a new reply at 10:00 and 2:00](https://opendoorlearningai.com/newsletter/lesson-5/3-a-day-with-it.png)

Hourly uses more of your plan than once a morning. Start with mornings and add hours once you trust the replies.

## When it gets something wrong

Say so in the same chat:

That's wrong: \[what's wrong\]. Fix it, and update the Email Triage task so you don't do it again.

Three examples: "Priya Shah is always Reply needed." "Never write a reply about price." "Only show me For your information items from clients."

If it says it can't change the task, open Email Triage under Scheduled and add the line to its instructions yourself. You can pause it there too or change how often it runs.

## This week

For one week read your assistant's review before you open your inbox. Each time it misses, tell it.

**What kind of email would you never let an assistant answer?** [Email me your answer](mailto:jonathan@opendoorlearningai.com?subject=Lesson%205%3A%20the%20email%20I%27d%20never%20hand%20off)

Up next: the weekly tidy, a Friday check that keeps your to-do list honest.

If this was useful, forward it to one owner who is drowning in email. New lessons go out every Thursday and the Monday issue covers what's worth trying: [opendoorlearningai.com/newsletter](https://opendoorlearningai.com/go/newsletter/fwd-lesson-5)
