AI-Generated Substacks? Em Dashes on Newsletters Have Surged Nearly 6x Since 2021
Em dashes — the punctuation mark most linked to AI-generated writing — now appear nearly six times more often in top Substack newsletters than before ChatGPT launched. However, other words that lingui
Before ChatGPT, the average top Substack essay contained about one em dash per 1,000 words. So far in 2026, it contains nearly five.
That is a 5.7-fold rise, and the strongest upward trend of any AI-writing marker I tracked in an analysis of 1,276 essays across 143 Substack publications and every calendar year from 2021 through April 2026.
The em dash is the punctuation mark most popularly recognized as an AI-writing fingerprint, though no peer-reviewed study has yet established it as a reliable AI detector on its own — and most stylometric researchers treat punctuation as one ingredient in a multi-feature ensemble, not a stand-alone tell. OpenAI itself, however, has acknowledged the tendency. Last November, CEO Sam Altman announced that GPT-5.1 had been tuned to suppress em dashes when users asked it to, a public concession he called a “small-but-happy win.” The accompanying post from OpenAI, in which ChatGPT “apologizes for ruining the em dash,” is about as direct as a vendor admission gets.
