What’s happened? One of the biggest problems with ChatGPT has now been fixed. Sam Altman announced via a X post that ChatGPT will now comply when users explicitly instruct it not to use em-dashes in the custom instructions tab. By adding a rule to the custom instructions to avoid using em-dashes, one can finally get ChatGPT to stop using them.
- The update addresses a long-running complaint that ChatGPT’s heavy reliance on the em-dash made its output look “bot-written.”
- The change works only when the preference is set in the Personalization or Custom instructions settings; telling it verbally in a prompt may not guarantee compliance.
- Altman said this is part of a broader push for customisation: ChatGPT already supports personality presets and memory features, and this is one of the earliest “style” controls rolled out.
Why this is important: Punctuation might seem negligible, but it plays a crucial role in writing tone, readability, and professionalism. For many users, such as writers, students, and professionals, the em-dash became a flashing signal that a text was AI-generated, and that undermined trust in ChatGPT’s usefulness for “human-style” writing. This update signals that OpenAI understands users don’t just want correct content; they want style control. Being able to tailor how ChatGPT writes (tone, slang, punctuation) brings it closer to a true writing assistant rather than a generic text generator.
That said, the fix also raises bigger questions: if the model struggled for years with one punctuation mark, what other stylistic quirks remain? And how many of those will require user-set rules rather than inherent improvement in the AI? This could be a step toward deeper AI personalisation, but it also shows how behavioural quirks linger in large-language models.
Why should I care? If you use ChatGPT for writing anything important, such as emails, reports, scripts, essays, or social posts, this update gives you more control over how the AI sounds, not just what it says. Style consistency matters, and removing obvious “AI tells” like the overuse of em-dashes can make outputs feel more natural and more you. It’s a small tweak, but it signals a bigger shift toward letting users shape their AI’s personality.
- It reduces the amount of editing you need to do by eliminating one of ChatGPT’s most recognizable writing quirks.
- It brings the AI closer to behaving like a true writing assistant you can customize, rather than a one-size-fits-all generator.
- It hints at more granular personalization tools in the future, giving you greater influence over tone, formatting, and stylistic habits.
Okay, so what’s next? Now that ChatGPT can quit its em-dash habit on command, the real question is how far this new obedience will go. If OpenAI keeps pushing in this direction, we may soon see deeper style controls: everything from sentence length to emoji tolerance to whether the AI should sound like a calm assistant or someone who’s had three coffees too many. This update also puts the spotlight back on personalization.
As such, you can expect creators, students, and businesses to test just how configurable the model really is, and whether these small switches can add up to an AI that finally blends into your writing instead of standing out. It’s early days, but if ChatGPT keeps taking notes this well, your future drafts might look a lot more “you” and a lot less “AI”.
