The em dash, this false proof of a text written by artificial intelligence (By Cheikh Tidiane NDIAYE) October 17, 2026
Can a hyphen betray the use of artificial intelligence in a text? On social networks, some readers seem to be convinced of this. The em dash (—), this long horizontal line which punctuates more and more publications, has become for some an almost infallible indicator of the intervention of ChatGPT or other conversational assistants.
However, this interpretation confuses a writing habit with proof of origin. In French typography, the hyphen is a punctuation mark which has several functions. It can introduce the lines of a dialogue, frame an incise or highlight an idea. It thus helps to organize the speech, to mark an interruption or to provide clarification. Its use is therefore not a recent invention of artificial intelligence, but rather typographical conventions established well before the appearance of generative tools.
Why, then, has this sign become so suspicious? The answer lies partly in the habits of language patterns. By learning from large volumes of text, including articles, essays, and published works, these tools reproduce stylistic regularities present in their training data. The em dash is one of these habits. Some assistants use it frequently to introduce an explanation, emphasize a contrast, or give rhythm to a sentence.
This trend is not just an impression. An analysis by the Pew Research Center published in August 2026, based on a sample of 490,000 texts from the Web, finds that the frequency of the em dash has approximately doubled compared to 2023. The researchers also note a more frequent use of this sign by artificial intelligence models. But they provide an essential precision: human authors also use this punctuation mark. Its presence therefore does not, in itself, allow us to conclude that a text was generated by a machine.
We must therefore distinguish two things: a statistical index and individual proof.
On the scale of a large set of texts, the frequency of certain signs or expressions can help identify stylistic developments associated with the use of AI. On the other hand, spotting an em dash in an article, a Facebook post or a professional message does not allow you to determine the author. A journalist who has mastered typographic rules can use it naturally. A writer can make it a characteristic of his style. A user can also keep it after having their text reread by a digital assistant.
The problem arises when this observation becomes a verdict. By associating certain signs with artificial intelligence, readers risk suspecting authentically human texts and confusing editorial quality, typographical correction and automation. The reverse is just as misleading: removing hyphens, multiplying mistakes or deliberately defacing a text does not guarantee that it will escape AI detection.
For journalists, teachers and information professionals, the stakes are high. Evaluating the origin of content cannot be based on a single punctuation mark. It requires a more global approach: examining the production context, comparing information with sources, analyzing possible inconsistencies and, when possible, focusing on the different stages of writing.
Artificial intelligence is transforming writing practices, but it has not invented the rules of language. The em dash may be an indicator of an evolution in writing habits in the era of generative models. It is not an exclusive signature of these tools.
When it comes to information, one rule remains essential: a suspicion is not proof. And a simple hyphen should never be enough to condemn a text.
