How to tell if an email was written by AI

Wikipedia's guide to AI text quotes this line as the opening of a machine-written request. In a cold email it spends the first line, the one the inbox preview shows, on nothing.

Instead: Open with the reason you are writing to this person.

Flattery that fits any recipient. Drafting tools reach for it when they have no real fact about the person.

Instead: Name the one specific thing you saw and why it made you write.

A scene-setting opener that says nothing about the reader. It reads as filler before the real point.

Instead: Delete the sentence and start with the second one.

Wikipedia's guide lists fill-in-the-blank templates that nobody filled as a sign of chatbot text. In an email you received, a raw blank means a template or a generated draft went out unfinished.

Instead: If this is your own draft, check that your sending tool fills every tag for every contact.

These are a chatbot talking to the person who asked it to write, pasted into the email by mistake.

Instead: Remove anything addressed to whoever wrote the draft rather than to the reader.

A model flags the gaps in what it could find. A person who researched you would just say what they know.

Instead: State the fact you have, or leave the point out.

In a study of over 15 million research abstracts, "delves" turned up 28 times as often in 2024 as earlier years predicted, and Wikipedia's guide lists both.

Instead: Say "look at", "go through" or "talk about".

The same study of research abstracts found "underscores" and "showcasing" far more often in 2024 than earlier years predicted (13.8 and 10.7 times), and Wikipedia's AI vocabulary list has all three.

Instead: Say "shows" or "points to", or cut the clause.

Both sit on Wikipedia's AI vocabulary list. Nobody calls a supply chain a tapestry in a sales email.

Instead: Say what the thing is.

Words a model reaches for far more often than people do. Wikipedia's guide lists each one.

Instead: Use the plain word, or a number that shows it.

Wikipedia lists these as AI vocabulary. In a cold email they sound like a report, not a person.

Instead: Say "build", "help" or "get".

Wikipedia's guide lists it among the words newer models overuse.

Instead: Say "use".

Wikipedia names these in its superficial analysis pattern. They claim a fit without saying what it is.

Instead: Say what you would change for them, in one concrete line.

Wikipedia's guide lists these under undue emphasis on significance. They make a small point sound grand.

Instead: Say what happens and to whom.

Wikipedia calls this a negative parallelism and lists it as a common sign of AI text.

Instead: Say the second half on its own.

Wikipedia lists this "-ing" tail as superficial analysis. It adds a claimed benefit without a fact behind it.

Instead: End the sentence before the comma, or make the benefit its own sentence with a number.

Wikipedia notes models avoid a plain "is" and use these instead.

Instead: Write "is".

Wikipedia lists the rule of three ("adjective, adjective, adjective") as a sign of AI text. One list is normal; two or more in a short email reads generated.

Instead: Keep the one item that matters most to this reader.

Wikipedia lists vague attributions as a sign of AI text. No reader can check who said it.

Instead: Name the source, or drop the claim.

Wikipedia lists this brochure tone as a sign of AI text from 2023 to 2025. Newer models rarely write it unless asked, so it now points to an older tool or a template.

Instead: Replace each adjective with the fact behind it.

Stock sales words that drafting tools reach for. Wikipedia's essay on marketing buzzspeak names "cutting edge" as one. They describe nothing the reader can check.

Instead: Say what changes for the reader, with a number if you have one.

Wikipedia lists the overused em dash as a sign of AI text from 2022 to September 2026. Newer models are often set never to use one, so an email without dashes proves nothing. Two or more in a short email still reads as generated to many readers.

Instead: Use a comma or a full stop.

Wikipedia lists overused boldface and markdown as signs of chatbot text. In a plain email they arrive as asterisks and hashes.

Instead: Write plain sentences.

Wikipedia lists inline-header vertical lists as a sign of AI text. People rarely format a cold email as a feature sheet.

Instead: Pick the one point that matters and write it as a sentence.

Wikipedia lists emoji placed at the start of headings or list entries as a sign of AI text.

Instead: Drop the emoji, or keep one in the whole email at most.