Guide7 min read·October 30, 2026

How to Avoid Sounding Like AI When You Write (Even If You Use ChatGPT)

People spot AI writing in a sentence or two, and it is rarely one word that gives it away. It is the stock opener, the "it's not X, it's Y", the three fragments in a row, the even rhythm. The tells to cut, what to write instead, and a two-minute edit for any draft.

PT
Presend TeamProduct
Watch: How to avoid sounding like AI when you write (even if you use ChatGPT) · 50 seconds

Why people notice

Most people now read AI-written text every day, in their inbox, on LinkedIn and in Slack. They have learned its habits without trying. When your message has those habits, the reader stops reading it as yours. They skim it, trust it a little less, and sometimes say so.

The good news: the habits are few and easy to see once you know them. You can use AI for a first draft and still sound like yourself, if you edit for these.


The tells, and what to do instead

1. The stock opener

"I hope this email finds you well." "I wanted to reach out regarding..." "Thank you for reaching out."

Instead: start with the point. "The contract is ready to sign." If a greeting matters to you, make it specific: "Good to see you at the offsite."

2. The stock closer

"Please don't hesitate to reach out." "I look forward to hearing from you." "At your earliest convenience."

Instead: end with the next step, or just end. "Can you sign by Friday?" Or "Thanks, Alex."

3. Model vocabulary

Delve, tapestry, landscape, realm, a testament to, leverage, utilize, facilitate, robust, seamless.

Instead: the plain word. Use, help, strong, smooth. Or nothing at all: "navigating the complexities of the market" usually means "selling".

4. "It's not X, it's Y"

"It's not just a tool, it's a new way of working." "This doesn't mean we stop. It means we focus." AI writes this shape constantly.

Instead: say Y. "It changes how we work." The contrast with a thing nobody said adds nothing.

5. Three fragments in a row

"No guessing. No waiting. No surprises." "Faster. Simpler. Cheaper."

Instead: one real sentence with the most important of the three, and a fact behind it. "You see the price before you book."

6. The staged set-up

"Here's the thing:" "Let's dive in." "Let's break it down." "Real talk:"

Instead: delete the run-up and keep what follows. It is almost always a stronger opening on its own.

7. The dramatic closer

"Let that sink in." "Read that again." "And that changes everything." "That's the real win."

Instead: stop one sentence earlier. If the point needed a drumroll, make the point better.

8. Inflated significance

"A pivotal moment." "Plays a crucial role." "Setting the stage for..."

Instead: say what happened and let the reader decide whether it was pivotal. "We hired our first engineer in Berlin."

9. Machine-even rhythm

Every sentence about the same length, every paragraph three sentences, no contractions. It reads smoothly and sounds like nobody.

Instead: mix a long sentence with a short one. Use "it's", "don't", "we'll". Let one paragraph be a single line.

10. The leftovers

"Certainly!" "Great question!" "Here's a revised version:" "Let me know if you'd like me to make it more formal." These are the chat window, pasted in by accident. So are hidden characters and a sudden run of em dashes. We wrote about both in hidden characters in ChatGPT text and why em dashes make email look AI-written.


A two-minute edit for any draft

Whether you wrote it or a model did:

1. Cut the first sentence. Read the draft without it. Usually better.

2. Cut the last sentence. Same test.

3. Add one detail only you would know. A name, a number, a date, something from the last conversation. Nothing makes text sound human faster.

4. Swap the long words. Leverage, utilize, facilitate, ensure. Read each and ask what you would say out loud.

5. Use contractions. Unless you are writing a contract.

6. Read it aloud. Anywhere you would not say it that way, change it.


If you draft with ChatGPT

  • Write the rough version yourself first, even three bullet points, and ask the model to tidy it rather than write it. The facts and the order stay yours.
  • Give it a sample of your writing and ask it to match the length and tone, not to "make it professional". Our guide to getting ChatGPT to write emails that sound like you has prompts that work.
  • Ask for shorter than you need. Models pad. It is easier to add one sentence than to cut five.
  • Always do the edit above. A model that is told "don't sound like AI" still does most of these things.

Check it before it goes

Paste any draft into our free AI slop detector. It underlines each tell, names it, and suggests a plainer swap where there is one. No sign-up.

To get the same check every time you send, Presend runs it in Gmail, Outlook on the web, Slack, LinkedIn and other places you write in the browser. It flags each tell by name, such as Stock opener, Model vocabulary or Dramatic closer, with a Fix button where there is a safe swap. Its voice check learns from your own sent messages and tells you which lines don't sound like you. For the phrase-by-phrase list, see words and phrases that make writing sound like AI.

Ready to try Presend?

Free to start. Nothing you write is stored. 30-second install.

Add to Chrome (free)
Related Articles
6 min read

How to Get an "Are You Sure?" Prompt Before Sending an Email Outside Your Company

No major email app stops and asks before an email goes to someone outside your company. Gmail, Outlook and Apple Mail only let you take it back for a few seconds. What each one gives you, how to get a real confirmation at Send, and how to keep people from clicking through it.

5 min read

How to Check Attachments Before Sending an Email (Missing and Wrong Files)

Gmail, Outlook and Apple Mail can all remind you when you write "attached" and forget the file. None of them checks that you attached the right file, which is the mistake that turns into a data breach. What each app catches, the habits that catch the rest, and how to add a check at Send.

7 min read

How to Stop Sending Sensitive Data by Email: DLP, Keyword Scans and Checks at Send

Card numbers, passwords and client details get emailed by accident every day. Microsoft 365 and Google Workspace can scan outgoing mail on some plans, Outlook add-ins like SafeSend scan for keywords, and a check at Send catches it while you can still fix it. How each works and which one you can set up today.

5 min read

Reply-All and Cc Mistakes: How to Stop Internal Emails Reaching Outside People

The worst wrong-recipient emails are replies: a reply-all with the client still on Cc, or a last-second Cc that autocomplete filled with an outside contact. Settings in Gmail, Outlook and Apple Mail that help, habits that help more, and a check that looks at recipients at the moment you click Send.

6 min read

Words and Phrases That Make Your Writing Sound Like AI (and What to Write Instead)

Delve, tapestry, leverage, "I hope this finds you well", "Let that sink in". The words and sentence shapes that make writing read as machine-made, grouped by type, each with a plainer replacement. Plus why one word is not proof, and what is.

6 min read

How to Humanize AI Text Without an AI Humanizer (5-Step Edit)

AI humanizer tools swap words for odd synonyms and often make text read worse. A five-step edit that actually makes an AI draft sound like a person wrote it, with a before and after, plus the prompts that give you a better draft to start from.