OpenAI Adds Finesse: How ChatGPT Is Learning to Drop Its Typographic Tell

What do 19th-century novels, eagle-eyed readers, and modern AI have in common? The answer lies in a subtle piece of punctuation – the em dash. This typographic flourish (—) has quietly become a hallmark of AI-generated writing. But now, OpenAI is turning the page: with ChatGPT 5.1, the em dash can finally be switched off – if you ask it to.

OpenAI CEO Sam Altman announced the tweak with evident pride on X: “Small but nice win,” he wrote. “If you tell ChatGPT not to use dashes in your custom instructions, it finally does what it’s told.” A throwaway remark, perhaps, but one that reveals just how far generative AI has come in refining tone and personalisation.

For years, models like ChatGPT developed what some users viewed as a textual fingerprint – a tendency to lean heavily on the em dash, more than human writers typically would. It was never a programmed feature, but a side effect of the model’s training data. Much of it drawn from older sources – including 19th-century literature – where the em dash was stylistically popular. Over time, this quirk became a statistical default.

And that’s the crux of the problem. Language models don’t “understand” instructions as humans do. They don’t recognise commands as rules – only as influenceable context. So when a user asked for no em dashes, the model might oblige briefly, but its training bias would often creep back in.

With version 5.1, OpenAI seems to have rebalanced that tension. Custom instructions are now more strongly prioritised, giving users finer control over how the model expresses itself. ChatGPT is learning – in its own way – to exercise restraint.

It may sound like a minor fix. But it reflects a deeper evolution in how AI writes: moving from broad generalisation to stylistic precision. And it matters more than it seems. Because in writing, it’s not just what is said that counts – it’s how. Dropping the em dash might look like a small gesture. But it symbolises something bigger: a shift away from generic AI output towards truly personalised language intelligence.

Alexander Pinker
Alexander Pinkerhttps://www.medialist.info
Alexander Pinker is an innovation profiler, future strategist and media expert who helps companies understand the opportunities behind technologies such as artificial intelligence for the next five to ten years. He is the founder of the consulting firm "Alexander Pinker - Innovation Profiling", the innovation marketing agency "innovate! communication" and the news platform "Medialist Innovation". He is also the author of three books and a lecturer at the Technical University of Würzburg-Schweinfurt.

Ähnliche Artikel

Kommentare

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Follow us

FUTURing