Economist Study Finds Sparse Punctuation Betrays AI Writing

Em dashes have become a popular shorthand for spotting machine-generated prose as generative AI spreads through publishing, education and business. But the punctuation mark alone is an unreliable signal, according to an analysis by The Economist. The more telling patterns are sparse punctuation, extended sentences and verbose phrasing — habits that may help distinguish output from large language models while also influencing how human writers shape their work.

The Economist examined more than 1.2 million words and compared text produced by leading models including ChatGPT, Claude and Gemini. Its findings showed that em dashes were not definitive evidence of AI authorship, while unusually low punctuation density and longer, more cumbersome sentences appeared more consistently across model output. The report also highlighted a feedback effect: writers are deliberately changing vocabulary and punctuation choices to avoid having their prose mistaken for AI-generated text.