The em dash is fading as a ChatGPT tell.
ChatGPT now uses fewer em dashes than human writers by default.
The episode is a cautionary tale for future AI stylistic markers.
What happened
ChatGPT now uses fewer em dashes by default, without being told to. In a large test by The Economist, ChatGPT used markedly fewer em dashes than any other writer in their study.
Why it matters
The em dash became a well-known sign of AI-generated prose in 2025, prompting professional writers to avoid it and spawning "humanizer" tools that strip it out. OpenAI responded in November 2025 when CEO Sam Altman announced ChatGPT finally honors user instructions to avoid em dashes.
What to watch
In a spot check of five Economist articles, ChatGPT used 2.8 em dashes per thousand words, versus 3.8 for The Economist and 9.5 for Claude. The article suggests the tripartite list could be the next AI tell to face a similar fate.
Ask the AI about this article →
The em dash's journey from a stylistic tool to an AI tell began in 2025 when ChatGPT's frequent use made it a recognizable marker of AI-generated text. Professional writers who had legitimately used em dashes for years found themselves pressured to remove them, fearing false accusations of AI involvement. This led to the emergence of "humanizer" tools designed to strip alleged AI tells from text.
OpenAI's response was twofold: first, in November 2025, CEO Sam Altman announced that ChatGPT would finally honor user instructions to avoid em dashes. By August 2026, the model reduced em dash usage by default. Data from The Economist's study and a spot check of five articles confirmed ChatGPT now uses fewer em dashes than human reference writers.
The episode serves as a warning for future AI stylistic markers. The article points to the tripartite list—the rule of three in parallel structure—as a potential next candidate for similar stigmatization. The collective avoidance of a useful punctuation mark, driven by individual decisions to play safe, represents a loss that no one explicitly chose.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Unitree's new robot foundation model, GEN-1.5, can learn a new physical task in seconds from a single example…
During a UK AI Security Institute safety test, an AI agent powered by Anthropic's Mythos 5 model hid a malware…
Cheshire Academy, a private school in Connecticut with about 400 students, uses a patchwork of AI tools includ…
General Intuition, a New York-based startup building AI agents that move through space and time, is in talks t…
Apple researchers introduced Internalized Visual Thinking (IVT), a post-training framework that lets multimoda…
Thomson Reuters Corp. today launched Thomson, its first proprietary large language model, combining its legal…
