OpenAI says it’s fixed ChatGPT’s em dash problem (techcrunch.com)

OpenAI says ChatGPT will now stop inserting em dashes when users explicitly tell it not to. The so‑called “ChatGPT hyphen” — a distinctive em dash usage that cropped up in essays, emails, posts and customer chats and became a popular (if unreliable) signal of AI‑authored text — drew criticism and mockery. CEO Sam Altman posted that if you add “don’t use em‑dashes” to your custom instructions, the model will “finally do what it’s supposed to do,” calling the change a “small‑but‑happy win.” OpenAI’s notes and a tongue‑in‑cheek Threads post underscore that this is a personalization improvement, not a default removal of em dashes. Technically, the update reflects tighter adherence to user prompts and customization settings: the model’s instruction‑following behavior has been tuned so style constraints in the personalization panel are respected more reliably. For practitioners and institutions this matters because it reduces a trivial stylistic fingerprint that fueled false positives in rudimentary AI detection, and it demonstrates progress toward controllable text generation—useful for editorial standards, brand voice, and academic integrity workflows. The caveat: users must opt in via custom instructions; the em dash won’t be universally eliminated unless explicitly requested.