Eased into AI
/ Learn AI From Zero / How to Use ChatGPT to Write Email in Three Passes
Learn AI From Zero 11 min read

How to Use ChatGPT to Write Email in Three Passes

Learn how to use ChatGPT to write email with a three-pass method for facts, tone, and final review, plus a reusable prompt and a realistic worked example.

An email draft moving through facts, tone, and final review cards

If you are learning how to use ChatGPT to write email, give it the confirmed facts first, ask for a draft with specific tone and length constraints, then review every name, date, promise, link, and attachment yourself. Copy the approved text into Gmail or Outlook only after that final check.

This guide uses three distinct passes for the job. They are facts, tone, and send-check. Keeping them separate stops a smooth first draft from hiding an invented detail. It also makes the result sound more like the sender because tone is edited after the message is accurate.

You can use this method in any email service. ChatGPT does not need access to your inbox for the basic workflow.

The Three-Pass Email Method

Pass Input Output Human Responsibility
Facts Audience, purpose, confirmed details, required action A complete factual draft Check every claim against your notes
Tone Desired relationship, directness, length, phrases to avoid A version that better fits your voice Rewrite anything you would not say
Send-check Final email and a fixed checklist A list of possible faults Make corrections, add attachments, and press Send yourself

The model never gets the final action. It helps shape text. You decide what another person receives.

I keep that boundary visible because email is deceptively consequential. A draft can look like ordinary office prose while quietly moving a deadline, promising an attachment that does not exist, or making a warmer relationship claim than the sender intended, and none of those mistakes becomes harmless because the paragraph is polite.

Pass 1: Give ChatGPT a Fact Sheet

Do not begin with “Write a professional email.” That request leaves the subject, recipient, purpose, facts, and expected response open to guesswork.

Write a tiny fact sheet first.

  • Recipient: who they are in relation to you
  • Purpose: the one outcome the email should achieve
  • Confirmed facts: names, dates, amounts, status, and previous commitments
  • Request: what you need the recipient to do
  • Deadline: only if one genuinely exists
  • Unknowns: anything the draft must not fill in

Use roles instead of real names while practicing. Remove account numbers, private customer details, health information, credentials, and confidential workplace material. OpenAI’s current Temporary Chat guidance says account holders can change whether conversations help improve models and that Temporary Chats are not used for training. Those controls do not make sensitive content necessary for a writing exercise.

Then use a prompt like this.

Draft an email using only the fact sheet below. Do not invent a reason, deadline, attachment, or previous conversation. If a required detail is missing, put it in a “Questions Before Sending” list after the draft. Keep the email under 140 words.

Paste the fact sheet beneath the request.

The questions list is useful because it gives uncertainty somewhere to go. Without that instruction, a model may produce a complete-looking email by smoothing over the gap.

A Worked Example With Fictional Details

This example is invented for the tutorial. It is not a message Kevin sent or a result from a real client relationship.

The fact sheet says the following.

  • Recipient is a freelance designer working on a landing page.
  • The agreed first draft date was September 8.
  • The writer needs to move the review call from September 9.
  • Tuesday afternoon or Thursday morning would work.
  • The new review date has not been chosen.
  • No extra work or payment change is being requested.

The first prompt could be this.

Draft a short email to the freelance designer. Explain that I need to move our September 9 review call, while the September 8 first-draft date stays unchanged. Offer Tuesday afternoon or Thursday morning for the rescheduled review. Do not invent a reason for the change. Do not imply extra work or a payment change. End with one clear question.

A factual draft might read as follows.

Subject: Moving Our Landing Page Review

Hi,

I need to move our September 9 landing page review call. The September 8 date for the first draft stays the same.

I can meet Tuesday afternoon or Thursday morning for the rescheduled review. Would either option work for you?

Thanks, Kevin

The email is plain, which is fine at this stage. Check it against the fact sheet before polishing anything. Both dates are present. The schedule change is separate from the draft deadline. No reason, fee, or extra deliverable appeared.

Pass 2: Shape the Tone Without Changing the Facts

Tone instructions work best when they describe a relationship and observable writing choices. “Make it better” is too vague. “Warm, direct, and without corporate phrases” gives the model something testable.

Continue in the same chat.

Keep every fact unchanged. Make the email warmer because we have worked together before. Use contractions, apologize once, and avoid “I hope this email finds you well,” “kindly,” and “at your earliest convenience.” Keep the body under 100 words.

Now read the version aloud. If you would never say a phrase in conversation, rewrite it yourself. Your normal sentence length, greeting, and sign-off do more for voice than asking the model to “sound human.”

Do not ask ChatGPT to imitate a named living writer or paste a private email chain to copy another person’s style. Describe the qualities you want. “Calm, concise, no exclamation marks” is enough for many work messages.

The fuller guide to writing good prompts explains why audience, task, context, and format matter across more than email.

Pass 3: Run a Send-Check

Paste the final draft back with a checklist request. This is not independent verification, because the same system is reviewing text it helped create. It is a way to surface obvious risks before your own inspection.

Review this draft without rewriting it. Make a table with one row for each name, date, number, commitment, requested action, link, and mentioned attachment. Quote the exact phrase and label it “confirmed from my fact sheet,” “not in my fact sheet,” or “needs my manual check.” Also flag vague deadlines and language that could sound blaming.

Then perform the real send-check yourself. I would not merge this with the tone pass, since reading for warmth puts the mind in a different mode from verifying whether Thursday is actually free and the correct person is in the recipient field.

  1. Verify the recipient and every copied address.
  2. Reopen your calendar for dates and time zones.
  3. Check every number, price, and commitment against the source.
  4. Open each link.
  5. Attach every file mentioned in the body.
  6. Remove comments, placeholders, and the model’s notes.
  7. Read the subject and first line together.
  8. Confirm that the requested response is clear.
  9. Read the whole email once from the recipient’s point of view.
  10. Press Send only when you own every sentence.

OpenAI’s current accuracy guidance warns that ChatGPT can produce incorrect facts and fabricated references. The send-check is where that general warning becomes concrete. A polished wrong date is still a wrong date.

A Reusable Email Prompt

Copy this structure and replace the brackets.

Write a [length] email to [recipient relationship]. The purpose is [one outcome]. Use only these confirmed facts: [facts]. Ask the recipient to [action]. The tone should be [two or three observable qualities]. Preserve [important wording or limitation]. Do not invent [unknown details]. If something necessary is missing, list questions after the draft instead of filling the gap. Include [subject line or other format requirement].

For a shorter rewrite, supply your existing draft instead of asking the model to reconstruct the whole story.

Shorten my draft to 120 words. Preserve every date, commitment, and question exactly. Remove repetition and formal filler. Do not add a subject or new claim.

The word count is an editorial constraint, not a guarantee that shorter is always better. A sensitive message may need enough space to avoid sounding abrupt.

Using ChatGPT With Gmail or Outlook

You do not need a browser extension, plugin, or inbox connection. The safest beginner route is manual.

  1. Draft in ChatGPT using sanitized facts.
  2. Complete the three passes.
  3. Copy only the approved email body.
  4. Paste it into Gmail or Outlook.
  5. Restore necessary real names and details from your trusted source.
  6. Run the send-check inside the email client.

Manual transfer creates a useful pause. You can see the actual recipient, thread history, attachments, and signature before sending.

That pause is not wasted efficiency. I would remove it only after a mature workplace process had replaced it with equally visible approval, source checks, logging, and permission controls, not because a plugin made the send button easier to reach.

Integrated AI features can reduce copying, but they also change what content the service can access and which workplace policy applies. Review the vendor documentation and your organization’s rules before connecting an inbox. Convenience does not settle the permission question.

Is It Okay to Use ChatGPT for Emails?

It can be okay to use ChatGPT for an email draft when you have permission to use the service, remove sensitive material, verify the content, and take responsibility for the message.

It is a poor fit when confidentiality rules prohibit sharing the content, the message gives consequential legal or HR notice, the recipient expects a personal response you are unwilling to write, or you cannot verify the claims. In those cases, write it yourself or use the qualified review process your organization requires.

Disclosure depends on context and policy. There is no universal rule that every AI-assisted sentence needs a label, and there is no universal permission to hide material AI use. Follow the rules governing the relationship and workplace.

Which AI Is Best for Writing Email?

There is no permanent winner. General assistants change, and email tasks vary. Compare tools with three sanitized examples from your own work.

Score factual fidelity, tone after one correction, amount of manual editing, privacy fit, and whether the free or paid access model matches your use. Do not let a tool win because its first draft used fancier language.

For most beginners, the tool already available is enough to learn the method. The facts, tone constraints, and send-check matter more than a feature leaderboard.

What Is the “Three Email Rule”?

There is no single official three-email rule for AI writing. The phrase is used for unrelated productivity and sales advice. The three passes in this article are an editorial workflow, not an industry law.

Use all three when a message contains commitments or could be misunderstood. For a casual note, the passes may take a minute. For a consequential work email, the final review should take as long as accuracy requires.

Common Failure Modes

Failure Why It Happens Correction
Invented deadline The prompt asks for urgency but supplies no date State “no deadline” or provide the confirmed one
Fake shared history “Make it warm” becomes an invented reference Ban invented previous conversations
Wrong attachment promise The draft assumes a file exists Mention only confirmed attachments and check them in the client
Excessive apology Tone request is vague Limit the apology and state the request clearly
Corporate filler “Professional” is interpreted as formal Name phrases to avoid and describe the relationship
Lost personal voice The whole email is accepted unchanged Read aloud and rewrite at least the lines that do not sound like you

The most dangerous failures are usually specific, not dramatic. A swapped date, hidden commitment, or missing attachment can survive because the paragraph around it sounds competent.

Keep the Final Inch Human

ChatGPT is good at turning structured facts into a draft and reshaping tone. It does not know whether the date in your notes is still valid, whether the recipient will read a phrase as blame, or whether you meant to promise the follow-up it added.

The broader routine for using AI in daily life follows the same boundary. Give the tool work you can inspect, then keep the decision and action under your control.

For email, the final inch is the entire point. Check the facts, make the language yours, look at the real recipient, and send only the version you are prepared to stand behind.