AI in PracticeOctober 8, 2026·5 min read

How to Use AI for Email Without Sounding Like AI

AI drafts emails fast but reads generic. Learn the brief that fixes it, the stock phrases to delete on sight, and which emails to keep in your own words.

"I hope this email finds you well." It shows up in inboxes constantly now, usually at the top of a draft nobody quite wrote. AI can produce an email in seconds; the hard part is making the result sound like a person who knows the reader.

The fix is mechanical: brief the AI with the email's job instead of the tone you want.

An email has a job

Nearly every email asks something of its reader: reply, approve, pay, show up, or take in a fact and move on. Ask an AI to "write a follow-up email" and it produces a follow-up-shaped object: polite, vague, and about nothing in particular. The reader may finish without knowing what you want, so the email misses the one thing it's for. The generic sound and the failure are the same problem: the draft fills space instead of doing anything.

Tone instructions disappoint for the same reason. "Make it warmer" rearranges the surface while the missing specifics stay missing, and the missing specifics are what the reader was looking for.

The brief

A working email brief contains four things: what happened, who's reading, what should happen next, and how you'd say it.

"Write a follow-up email to a client about a quote."

"Draft a reply to Dana about the quote we sent Tuesday. It's been five days. I don't want to nag, but the job can't be scheduled until she approves it. Three sentences, no exclamation marks, end with a clear question."

The second gives the AI what a colleague would need: the history, the stake, the ask, and the register. The first gives it a genre and hopes. The drafts show the gap:

Dear Dana, I hope you're doing well. I wanted to follow up regarding the quote we sent on Tuesday. Please don't hesitate to reach out with any questions. I look forward to hearing from you.

Hi Dana, checking in on the quote from Tuesday. I can't schedule the job until you approve it, so a yes or no by Friday would help a lot. Any questions, just reply here.

Same request. One fills space, the other does the job, and the difference is visible in one line.

The brief works as a fill-in prompt too:

Draft a reply to [name] about [thing].
Context: [what happened, in one line].
They should: [what you want them to do].
Length and tone: [N sentences, the way I'd say it out loud].

That last line does quiet work. "The way I'd say it out loud" is what turns "I wanted to circle back at your earliest convenience" into "could you let me know by Friday?"

Delete these on sight

Stock phrases are the tell readers catch first, and deleting them costs nothing.

Ordinary courtesies survive the cut: "Thanks, got it" is a complete email. The test we apply: would a person say the sentence out loud to this specific reader? One stock phrase reads as habit; six in a row read as a template.

Where AI earns its keep

AI helps most with the emails you keep not writing. The follow-up delayed for a week, the no that needs the door left open, the awkward ask: those stall because starting them is unpleasant, not because finishing them is hard. Hand over the situation and let the AI produce version one:

"Draft a reply declining the invitation. Thank them by name, one honest sentence about why it's a no this time, and leave the door open for the next one. Four sentences."

Condensing is the other quick win. Paste a four-paragraph draft and ask for three sentences. The short version usually needs your edit, not a rewrite.

Drafting is also built into the inbox now. Gmail's Help me write produces a draft from a prompt, offers one-click refines like Formalize and Shorten, and pulls details from your other emails and Drive files (Draft emails with Gemini in Gmail). ChatGPT holds standing preferences too: tell it once that your work emails get no exclamation marks, and drafts afterward start closer to you. How to Use ChatGPT Memory Effectively covers how to set that up.

Write these yourself

A few kinds of email are worth writing yourself, and they're easy to spot: something went wrong, something is contested, or someone will feel the result.

Apologies top the list. An AI apology reads like a press release because it optimizes for plausible wording, and accountability is specific: what went wrong and what happens next. Draft it with AI if the blank page stalls you, then write the accountability sentence yourself.

Negotiations come second. Salary, rates, vendor pricing: the number and the reasoning are yours, and a hedged draft is expensive in ways that are easy to miss.

Bad news about a person rounds it out: a rejection, a role change, a project ending. The reader may reread it more than once, and the words need to hold up.

The closer the relationship, the more a generic draft stands out. A colleague who emails you weekly is likely to feel a register shift before they can name it. For those threads, AI works better as an editor than a writer: paste your version and ask for a trim.

FAQ

Is it OK to use AI for work emails?

For routine email sent on your own behalf, it's a tool, like a colleague sketching a first draft. Two checks first: follow your employer's AI policy, and don't paste confidential or personal details into a chatbot whose data settings you haven't reviewed.

Can you tell if an email was written by AI?

Often, yes, but by reading, not by software. Detection tools are unreliable in both directions, and the signal a human picks up is simpler: stock phrases, zero specifics, no dates or names. An email with your details in it stops scanning as machine-made.

Which AI should I use for email: the one in Gmail or ChatGPT?

Whichever is open when you're writing. Built-in drafting knows your inbox context; ChatGPT handles the blank-page emails and holds standing preferences across tools. The skill transfers either way: brief with the job, then edit.

Do I have to tell people an email was AI-drafted?

For ordinary work email, disclosure expectations depend on your employer, your clients, and the context of the message. Check those where they exist. For everything else, the question worth more attention is whether the email is accurate and does its job.


Want to go further?

Want to keep going?

Get the next one delivered.

If you liked this, the next one is worth an inbox slot.