The Em Dash is Gone: Sam Altman Confirms ChatGPT Finally Dumps Its Most Annoying Habit

OpenAI CEO Sam Altman has confirmed via his personal X account that ChatGPT has addressed one of its most frequently criticised “bad habits.” According to OpenAI, if users specify in Custom Instructions that they do not want ChatGPT to use the em dash, the model will better respect that preference, reducing one of the more obvious telltale signs of AI‑generated text. Previously, prompts could be ignored and ChatGPT continued to rely heavily on em dashes, a feature that many used as a heuristic for AI authorship.

A prevailing theory suggests the punctuation arises from training data, as large language models are trained on vast datasets where em dashes appear frequently, and the training process does not flag this punctuation for limitation. With the latest adjustment, it will be more difficult to judge AI authorship based solely on em dashes, meaning readers may need to scrutinise online content more carefully. The post, dated 16 November 2025, notes that this change aims to reduce one easily observable AI-writing quirk.