ChatGPT Words to Avoid in Email, and What to Say Instead

Why some words give AI away

Language models learned to write from a huge amount of published text, and they lean on a small set of words far more than people do when they type an email. Researchers have measured it. A study of more than 15 million biomedical abstracts, published in Science Advances in July 2025, found that words like "delves", "showcasing" and "underscores" jumped in 2024, right after ChatGPT came out. In the preprint, "delves" showed up about 28 times more often than the trend predicted. The authors estimate that at least 13.5% of 2024 abstracts were processed with a language model.

Email readers have learned the same pattern without any data. One of these words in an email proves nothing. Three of them in a short reply, and the reader stops believing a person wrote it.

Here are the words and phrases to watch for, grouped by where they show up, with a plain swap for each.

Single words

Most of these have a shorter word that means the same thing. When there is no swap, the word is usually padding and the sentence works without it.

Openers

The first line is where AI drafts are easiest to spot, because the model fills space before it gets to the point.

• "I hope this email finds you well."
• "I hope you're doing well."
• "I wanted to reach out regarding..."
• "Thank you for reaching out."
• "I trust this message finds you..."

Instead, start with the reason you are writing: "I'm writing about the March invoice." "Quick question about Friday's call." If you want to be warm, be specific: "Good to see you at the offsite."

Closers

• "Please don't hesitate to reach out if you have any questions."
• "I look forward to hearing from you."
• "At your earliest convenience."
• "Please feel free to..."
• "Let me know if you'd like me to..."

Instead, say what you need and when: "Can you confirm by Thursday?" If you have nothing to ask, "Thanks," and your name is enough.

Padding

• "It's worth noting that..."
• "I would like to take this opportunity to..."
• "In today's fast-paced world..."
• "Please be advised that..."
• "After careful consideration,..."

Instead, write nothing. Delete the phrase and start the sentence at the next word. "It's worth noting that the price goes up in May" becomes "The price goes up in May."

Sentence shapes

Swapping words is not enough if the shape of the sentence gives it away. People now recognise these on sight:

• "It's not just X, it's Y." Say Y.
• "No meetings. No delays. No surprises." A row of fragments reads like an ad. Write one sentence.
• "Here's the thing:" and "Let's dive in." These announce the point instead of making it. Cut them.
• "Let that sink in." and "Read that again." Nobody needs to be told to think.
• Everything in threes. "Fast, reliable and secure." If you have two points, write two.
• Long dashes. A typical AI draft has two or three em dashes per paragraph. We wrote a whole post on why.

Chat window leftovers

Some phrases only end up in an email because the text was copied straight out of a chat:

• "Certainly!" or "Absolutely!" as the first word
• "Great question!"
• "Here's a revised version:"
• "I'd be happy to help with that."
• "As of my last knowledge update..."

These are the worst ones to send, because they show the reader exactly how the email was made. Read the first and last lines of anything you paste. Copied text can also carry invisible characters that you cannot see in the compose box.

One word is fine. The pattern is the problem.

Plenty of people say "crucial" or "leverage" out loud, and nobody needs to rewrite an email to avoid one word. What readers notice is the cluster: an opener, a "delve", a list of three and a closing offer to answer any questions, all in six lines.

So when you edit, count kinds of tells, not words. One habit is your style. Three different tells in a short email is a draft that still reads like a chatbot wrote it. Our guide on making an email sound human walks through a full before and after.

Check a draft in seconds

Paste any draft into our free AI slop detector to see which of these it contains. To get the check on every email, Presend runs it when you click send in Gmail, Outlook, Slack and the other places you write. It underlines each phrase and suggests a plainer swap where there is one. It also counts different kinds of tells rather than single words, so one stray "leverage" gets a mild note, not a slop verdict.