How to Make ChatGPT Emails Sound Human: 7 Edits With Before and After Examples

Updated October 5, 2026

To make ChatGPT emails sound human, edit the draft after it's written. Delete the stock opener, swap vague claims for one specific detail, use words you'd say out loud, vary your sentence lengths, ask for one small thing, cut the recap and the closer, and read the result aloud. Seven edits, about five minutes per email. Changing the prompt helps, but the edits are what remove the polished-but-empty feel that readers spot fast. Below, each edit has a before and after.

The draft we'll fix

Here is a typical first-pass cold email from ChatGPT. The company and names are made up for illustration.

Subject: Exciting Opportunity to Optimize Your Workflow

Hi Dana,

I hope this email finds you well. I wanted to reach out because I came across your company and was impressed by your innovative approach. Our comprehensive platform streamlines invoicing, reduces errors, and saves valuable time. I would love to schedule a call to discuss how we can help you achieve your goals. Please don't hesitate to reach out with any questions. Looking forward to hearing from you!

Nothing in it is wrong. Nothing in it is about Dana, either. Any of the seven edits below would improve it; together they change what it is.

The 7 edits

1. Delete the opener

Drafts tend to start with a greeting ritual: "I hope this email finds you well," "I wanted to reach out." The reader skips it, and it tells them the rest is templated. Start with the reason you're writing.

Before: I hope this email finds you well. I wanted to reach out because I came across your company.
After: I saw Brightline posted two AP clerk roles this month.

2. Replace the vague claim with a specific one

"Innovative approach," "valuable time," and "achieve your goals" could be pasted into any email to anyone. Ask yourself what you actually know about this person or what your product actually does, and write that.

Before: Our comprehensive platform streamlines invoicing, reduces errors, and saves valuable time.
After: We match each invoice to its purchase order and flag the ones that don't line up before payment goes out.

The model can't supply this part. It doesn't know what you know. Paste in the facts (the job posting, the product detail, the thing you noticed) and edit around them.

3. Swap formal words for the ones you'd say

Drafts lean on words like "utilize," "facilitate," "comprehensive," and "innovative." Lists of overused AI words exist for a reason; we keep one here. Don't just ban words, though. Replace each one with what you'd say across a table. Use contractions: "don't," "we've," "it's."

Before: I would love to schedule a call to discuss how we can help you achieve your goals.
After: Could we talk for 15 minutes next week?

4. Break the rhythm

ChatGPT likes lists of three ("streamlines, reduces, saves") and sentences of similar length that march along evenly. People write unevenly. A long sentence with a clause or two, then a short one. A fragment, if it fits.

Before: We help teams save time, reduce errors, and improve accuracy.
After: We help teams catch invoice errors earlier. That's the whole pitch.

Also look for the "it's not just X, it's Y" construction and for em dashes stacked two or three to a paragraph. Rewrite them as plain statements.

5. Shrink the ask to one small thing

"Schedule a call to discuss" is a big ask for a stranger. Make the request easy to say yes to, and easy to decline. One question, one next step.

Before: I would love to schedule a call to discuss how we can help you achieve your goals.
After: Want me to send a two-minute screen recording? If not, no problem.

6. Cut the recap and the closer

Drafts end by restating the pitch, then add a polite flourish: "Please don't hesitate to reach out," "Looking forward to hearing from you!" Delete both. End on the ask, then your name.

Before: Please don't hesitate to reach out with any questions. Looking forward to hearing from you!
After: (Nothing. The email ended one line earlier.)

7. Read it aloud, then fix what you wouldn't say

This is the check that catches everything else. If you stumble, or feel slightly embarrassed saying a line, rewrite it. Do this as the sender, not as an editor. Your own tics are fine to keep: a short "Quick question," a dry aside, a word you always use.

The full email, after

Subject: Invoice errors at Brightline

Hi Dana,

I saw Brightline posted two AP clerk roles this month. Teams at that stage often start losing hours to invoice mismatches.

We match each invoice to its purchase order and flag the ones that don't line up before payment goes out. That's the whole pitch.

Want me to send a two-minute screen recording? If not, no problem.

Sam

It's roughly half the length. It says one thing, to one person, and asks for one small thing. For more on this format, see Plainspoke for cold email.

Fix the prompt too, so there's less to edit

You'll still edit, but a better prompt gives you a better starting point. Two approaches from outside sources are worth copying:

Describe your style from real samples. Zapier's walkthrough suggests using ChatGPT to analyze a few samples of your writing for voice, tone, and structure, then pasting those descriptors into custom instructions.
Choose the right level. OpenAI's customization guide says instructions can be set in your settings, in a specific project, or in the chat itself for a one-off change.

Give the model the reader, the goal, and the specific facts in the prompt. And be aware that a list of banned words only treats symptoms. Drafts that avoid every flagged word can still read as generic if every claim is vague. For the broader process, read how to make AI writing sound human.

Mistakes that keep emails sounding robotic

Swapping synonyms. Replacing "utilize" with "use" while leaving the vague claim is a cosmetic fix. Edit 2 matters more than edit 3.
Adding fake warmth. "I'm so excited to connect!" reads as a template too. Warmth comes from specifics about the reader.
Adding typos on purpose. Some forum advice says to do this. It makes you look careless without making you sound like yourself.
Editing for a detector instead of a reader. Detectors and humans react to different things, and detectors sometimes flag writing that people wrote. See why AI detectors flag human writing. Write for the person who gets the email.
Using the same fix on every email. A LinkedIn message, a resume bullet, and a follow-up each need different handling. We have pages for LinkedIn and resumes.

When a tool helps, and when it doesn't

You can do all seven edits by hand, and for a single important email you probably should. Where a tool earns its place is volume: when you send many emails a week and want something to flag the stock phrases, vague claims, and uniform rhythm so you only handle the judgment calls. That's what Plainspoke is built for: checking and rewriting professional writing such as cold email, LinkedIn, resumes, and product pages. It can't know the facts only you know, so edit 2 stays your job. Plainspoke is not meant for schoolwork. If you're comparing options, see the comparison pages to weigh Plainspoke against other tools, and check pricing before you decide.

Sources

Check a draft free

Frequently asked questions

Can people tell when an email was written by ChatGPT?

Often, yes. Not because of one word, but because of the pattern: a stock opener, vague praise, evenly sized sentences, and a polite closer that restates the pitch. An email with a specific detail about the reader and a short, plain ask rarely reads that way, whatever tool drafted it.

Should I tell ChatGPT to 'sound human' in the prompt?

It's too vague to help much. Say who the reader is, what you want from them, and how you actually write. Pasting two or three of your own emails as samples gives the model something concrete to match. Then still edit the output.

How long does it take to humanize an AI email by hand?

For a short cold email, about five minutes once you know the seven edits. The slowest part is finding the specific detail about the reader for edit 2, and that's also the part that matters most.

Is it dishonest to use ChatGPT for work emails?

Using a drafting tool isn't the problem. Sending claims you haven't checked is. If you read the draft, correct anything untrue, and put your own facts in it, the email is yours. Follow any rules your employer or industry sets on AI use.

Does this work for LinkedIn messages and resumes too?

The same principles apply: specific over vague, plain words, one ask. The details differ. LinkedIn messages should be shorter, and resume bullets should name results you actually achieved. See the LinkedIn and resume pages for specifics.