You need: a free account on ChatGPT, Claude or Gemini, and a notes app. Cost: nothing.
Before you paste anything
Check your workplace's AI rules first. Some employers don't allow work email in AI tools. If yours does, strip out what the AI doesn't need: write
[Client]instead of a name, and leave out account numbers, addresses, prices you haven't agreed and anything confidential. Also check your assistant's Settings: unless you switch it off, your chats may be used to train its AI.
Step 1: Write your email card (5 minutes)
Copy this into your notes app and fill in every bracket. This is what makes the drafts sound like you.
EMAIL REPLY CARD
You draft replies to my work emails. I'll paste an email and one line
on what I want to say. You write the reply. I check it and send it.
ABOUT ME
- Role: [e.g. office manager at a dental clinic]
- Who writes to me: [e.g. patients, suppliers, my manager]
- My tone: [e.g. warm, plain English, no jargon]
- Sign-off: [e.g. "Best, Sam"]
WHEN I PASTE AN EMAIL
1. If I haven't said what I want to reply, ask me in one short line
before writing anything.
2. Write one reply, under 120 words, ready to send.
3. Under it, add "CHECK:" and list any fact you assumed, such as a date,
a time or a promise, so I can confirm it.
RULES
- Never promise a date, price, refund or deadline I didn't give you.
- Never invent facts, names or attachments.
- Answer every question the sender asked, in the order they asked.
- If the email is angry, stay calm and polite, and don't admit fault
unless I tell you to.Step 2: Draft your first reply (3 minutes)
Open a new chat. Paste your whole card, then the email, then one line on what you want to say. For example:
Here's the email:
"Hi, could we move Thursday's meeting? Wednesday works for me. Also,
can you send the latest figures beforehand?"
What I want to say: yes to moving it, but Friday morning not
Wednesday; figures by Thursday evening.Read the draft and its CHECK list. Change anything that isn't right, then copy it into your email app.
Step 3: Test it (5 minutes)
Three quick tests show whether your card holds up:
- The vague test: paste an email with no instruction. It should ask what you want to say, not guess.
- The tricky test: paste a complaint, for example "This is the third late delivery. I want a refund." with "Apologise, say I'm checking with the warehouse." It should stay calm and must not promise a refund.
- The change test: reply
Shorter and warmer.It should rewrite the same reply, not start a new one.
If a test fails, don't start over. Add one rule to your card, such as "Never offer money or refunds unless I say so," and try again. Test, fix, retest: that's how a chat becomes a tool you trust.
Step 4: Keep it (2 minutes)
Your card is now a reusable tool. Next time, open a new chat, paste the card, then the email. If your assistant can save instructions (a Project in ChatGPT or Claude, or a Gem in Gemini), you can paste the card in once instead. Menus change, but the card in your notes always works. Hit a limit? Paste your card into another assistant and carry on.
Good habits
- You send, not the AI. Read every draft in full. The reply goes out in your name.
- Give it your facts. The more specific your one line ("Friday 10 am, figures Thursday"), the less it has to guess.
- Don't paste a whole thread if the last email is enough. Less pasted means less shared.
What this doesn't do
On a free plan, the assistant only sees what you paste. It doesn't read your inbox or send anything. That's a feature for privacy, and it keeps you in control. Sorting a whole inbox and drafting replies in bulk is a bigger build: it's one of the 12 tools in Capstone AI Lab, the email autoresponder, where you still review and send.