My cousin runs a small interior design studio in Pune, and a few months ago she started using ChatGPT to draft client emails. Within two weeks, a long-time client replied to one of them with a single line: “Is this you or your new assistant?” The email wasn’t wrong. It was polite, well organised, and full of phrases she would never say out loud, like “I hope this message finds you well” and “Please don’t hesitate to reach out.” The client had been getting her short, slightly chaotic WhatsApp-style emails for three years. The change stood out immediately.
Give an AI tool the facts and it will usually get them right in an email. What it can’t pick up on its own is the way you, specifically, talk to the person on the other end.
People who know you notice a voice shift faster than they’d notice a typo. Colleagues, clients, your manager, your landlord: they all have a sense of how you write, even if they couldn’t describe it. When an email suddenly sounds like a customer support template, they don’t think “how efficient.” They think something’s off.
What makes AI emails sound like AI
Before fixing anything, it helps to know what to look for. When I read AI-drafted emails side by side with the sender’s normal writing, the same differences show up again and again.
Openers and closers nobody uses. “I hope this email finds you well.” “I wanted to reach out regarding.” “Looking forward to hearing from you at your earliest convenience.” Some people really do write like this. Most don’t.
Even, medium-length sentences. AI drafts tend to have every sentence roughly the same length, around fifteen to twenty words. Real emails are lumpier. A short line. Then a longer one that wanders a bit because you’re explaining something while also remembering another detail.
Too much structure for a small message. A three-line request doesn’t need bullet points, a bold heading and a summary paragraph.
Over-politeness. Lots of “kindly,” “I would greatly appreciate,” “thank you so much for your understanding.” In Indian offices, “kindly” is common in human writing too, which makes this tricky. The giveaway is density: AI stacks three polite phrases into a two-sentence email.
No personal references. Your real emails mention the call from Tuesday, the chai meeting, the fact that the vendor was late again. AI drafts float free of any shared history unless you feed it in.
Step 1: Collect samples of how you write
This takes ten minutes and makes every later step work better. Open your sent folder and copy five or six emails you wrote yourself, without AI help, to different kinds of people: a client, your boss, a colleague, a vendor, maybe a slightly awkward one where you had to say no.
Paste them into a single note. Then read through and jot down a few things you notice. Do you open with “Hi” or “Hello” or just the person’s name? Do you sign off with “Thanks,” “Regards,” “Cheers” or just your initial? Do you use exclamation marks? Do you write in short paragraphs or one block? Do you mix in Hindi or another language with certain people (“Haan, will do” or “Let’s catch up post lunch”)?
That short list becomes your voice guide.
Step 2: Give the AI your voice before asking for a draft
Most people type “Write an email to my client about the project delay” and accept whatever comes out. Try this instead:
Here are three emails I wrote myself. Study the tone, sentence length, greetings and sign-offs.
[paste samples]
Now write an email in the same style to a client named Rohan. The kitchen cabinets will arrive five days late because the supplier’s truck broke down. New delivery date is the 18th. Offer to adjust the carpenter’s schedule at no extra cost. Keep it under 100 words. Don’t use “I hope this finds you well” or “please don’t hesitate.”
The banned-phrases line matters. Tell the model what you never say and you skip half the editing.
Some people prefer to work the other way round: get a quick generic draft, then pass it through an ai humanizer to loosen the stiff phrasing and even out that very regular sentence rhythm before they add their own details. Either order works. The samples approach gives you a closer match to your voice, and the humanizer route is faster when you’re clearing twenty routine replies before lunch.
Before and after: three real-style examples
All three of these are composites, but they’re the kind of emails I see people send every week.
A delay update to a client
AI default:
Subject: Update Regarding Your Kitchen Project
Dear Rohan,
I hope this email finds you well. I wanted to reach out to inform you that, due to unforeseen circumstances with our supplier, the delivery of your kitchen cabinets has been delayed by approximately five days. The revised delivery date is now the 18th.
We sincerely apologize for any inconvenience this may cause. To minimize disruption, we would be happy to adjust the carpenter’s schedule at no additional cost to you.
Please don’t hesitate to reach out if you have any questions or concerns.
Warm regards, Meera
After editing in her own voice:
Subject: Cabinets moved to the 18th
Hi Rohan,
Bad news on the cabinets, the supplier’s truck broke down near Nashik so they’ll reach you on the 18th instead of the 13th. Really sorry about this.
I’ll shift the carpenter to the 19th so there’s no gap, no extra charge for that. If the 19th doesn’t work for you, tell me and we’ll figure it out.
Meera
The second version is shorter, specific about the cause, and sounds like a person who has been dealing with Rohan for months. Notice the comma splice in the first line. It’s how she writes, so it stays.
Asking your manager for leave
AI default:
Dear Mr. Sharma,
I am writing to formally request leave from October 20th to October 24th to attend a family function. I have ensured that my current tasks are up to date and have coordinated with my team members to cover any urgent matters during my absence. I would greatly appreciate your approval of this request.
Thank you for your consideration.
Best regards, Arjun
After:
Hi Vikram,
Need leave 20 to 24 Oct for my sister’s engagement in Lucknow. The vendor dashboard handover is done and Neha has agreed to cover the Monday client call.
Let me know if that’s okay.
Arjun
If Arjun normally calls his manager by first name, the AI version’s “Mr. Sharma” would have looked strange. Small details like that give a draft away faster than any vocabulary choice.
Saying no to a vendor
AI default:
Thank you for your detailed proposal. After careful consideration, we have decided not to move forward at this time. We appreciate your time and effort and will keep your information on file for future opportunities.
After:
Thanks for sending this over so quickly. We’re going to pass for now, mainly because the budget for this quarter is already committed. If things change in January I’ll come back to you first.
The second one gives a reason, which feels respectful, and a realistic next step instead of the vague “keep on file” line that everyone knows means nothing.
Step 3: Edit for the five things that give it away
Once you have a draft, run through this short list. It takes under two minutes per email.
- Replace the opener. Delete any “hope this finds you well” line. Start with the actual point, or with whatever you usually start with.
- Cut one polite phrase per paragraph. If there are three, keep the one you’d really say.
- Add one specific detail only you would know. The supplier’s location, the name of the colleague covering for you, the date of your last call.
- Break the rhythm. Make one sentence much shorter. Let another one run on a bit.
- Swap the sign-off for the one you actually use.
If you’re editing a lot of drafts and the tone keeps coming back stiff, a humanizing tool can do a first pass on the rhythm and word choice. I’d still read the result and add your personal details by hand, because no tool knows that the cabinets got stuck near Nashik.
Step 4: Build a few reusable prompts
For emails you send often, write a saved prompt with your voice guide built in. A simple version:
You are drafting emails for me. My style: short paragraphs, no more than two lines each. I open with “Hi [name],” and sign off with just my first name. I don’t use “kindly,” “I hope this finds you well,” “please don’t hesitate,” or “at your earliest convenience.” I sometimes start sentences with “And” or “But.” I mention specific dates and names instead of vague references. Keep every email under 120 words unless I say otherwise.
Paste this at the start of each conversation, or save it in your AI tool’s custom instructions if it has that setting. Then you only need to type the situation: “Follow-up with Anjali about the invoice from 2 September, still unpaid, polite but firm.”
Where AI helps and where it doesn’t
| Email type | AI draft useful? | What to watch |
|---|---|---|
| Routine status updates | Very | Remove filler opener, add specifics |
| Formal letters (bank, government, HR) | Yes | Formal tone is fine here, check facts |
| Delicate conversations (conflict, bad news) | Partly | Rewrite the key sentences yourself |
| Quick replies to close colleagues | Rarely | Faster to just type it |
| Cold outreach | Partly | Generic drafts get ignored, personalise heavily |
For formal letters to a bank or a government office, the polished AI voice is often exactly right. Nobody at the passport office expects your personality. Where voice matters most is with people who already know you.
Common mistakes
Sending without reading aloud. Read the email under your breath before sending. If any sentence makes you wince, rewrite it. Your ear catches what your eye skips.
Over-correcting into fake casual. Some people, once they realise AI sounds stiff, ask it to “make it casual” and get something full of “Hey there!” and “Super excited to share.” That’s just a different template. Match your real level of casualness, which for most professionals is somewhere in the middle.
Letting AI add promises. I’ve seen drafts where the AI helpfully added “We will provide a full refund” or “This will not happen again” when the sender never intended to commit to that. Check every commitment in the email.
Using the same draft for different people. Your email to your CEO and your email to your work friend about the same meeting shouldn’t read identically. Adjust the greeting, length and tone per person.
Forgetting regional habits. Indian English has its own normal phrasings: “do the needful” (which many people now avoid), “revert back,” “PFA.” Decide which of these you actually use and tell the AI. If you don’t use them, ban them. If your client uses them and you mirror that, keep them.
For longer pieces, a monthly client newsletter for example, I usually draft with AI, then run the text through an ai humanizer to smooth out the uniform rhythm, then do a final read where I add two or three personal touches. That three-step routine takes maybe fifteen minutes for something that used to take an hour.
A few quick answers
Is it dishonest to use AI for emails? Most people consider it fine for drafting, much as using spell check is fine. It becomes a problem if the content misrepresents what you think or promises things you didn’t mean. You’re still responsible for every word you send.
Can people tell when an email is AI-written? Often, yes, especially people who know your writing. The tells are generic openers, even sentence lengths and the lack of specific shared context. Editing for those three things removes most of the signal.
Should I tell clients I use AI? There’s no rule. If a client asks, a straight answer is better than an awkward denial.
Try it on tomorrow’s first email
Before you open a chatbot tomorrow morning, pull three of your own sent emails and save them in a note called “how I write.” Use them in your first prompt of the day, then compare the draft with what you would normally have typed. You’ll start seeing your habits more clearly, and after a week or so you might find the AI draft needs only a minute of editing, or that for some emails you’re quicker typing them yourself anyway. Both outcomes are fine.
