Do em dashes mean AI wrote it?

No. The em dash is ordinary punctuation that writers and editors have used for centuries, and its presence tells you nothing about who — or what — produced a sentence.

The myth has a real root. ChatGPT's default style leaned on em dashes heavily enough that people started spotting them in colleagues' emails, and for a while "suspiciously many dashes" was a fair hunch about which tool someone had reached for.

The tell got switched off

In November 2025 OpenAI made ChatGPT obey a custom instruction to stop using them. Sam Altman announced it as a "small-but-happy win": tell it no em dashes, and you get no em dashes. A habit you can turn off in one line of settings is not evidence of anything. It stopped meaning much the moment it became famous enough to be worth suppressing.

It never worked in reverse either. People use em dashes because they were taught to, or because their house style prefers them to brackets. A model asked for plain prose writes plain prose, with none at all.

The same goes for the rest of the list

"Delve", "it's not just X, it's Y", the tidy three-bullet summary, the closing paragraph about what it all means — these are defaults, and defaults are the easiest thing in the world to change. Anyone deliberately passing off machine text has already changed them. The writers left holding the tells are the ones not hiding anything: people who like dashes, people who were taught to summarise, people who write carefully.

What can actually be measured

Invisible characters, sometimes. They survive copy-paste and you can check any text for them in a few seconds — but they show what software the words passed through, not who wrote them, and most text has none.

They are not a watermark from OpenAI. ChatGPT does not watermark its text output.