For years, one punctuation mark quietly gave away AI generated text. The em dash. Writers, students, and professionals often joked that ChatGPT could not resist using it. Now, OpenAI has taken that criticism seriously. Sam Altman confirmed in a post on X that users can finally train ChatGPT to avoid em dashes by setting a preference inside the Custom Instructions menu, which lives under the Personalization settings.
Altman explained that ChatGPT will follow this stylistic preference when users configure it inside their account settings. Telling the model directly inside a single prompt may not fully work yet, but the personalized rule appears to override its default writing habits. You can read Altman’s announcement.
This update resolves one of the most common complaints from long-time ChatGPT users who rely on the AI to draft emails, reports, scripts, academic work, and professional communication. The em dash habit often made content feel “AI written,” even when the ideas were strong. Many editors and marketers recognized the pattern instantly. That tiny stylistic quirk chipped away at the credibility and authenticity of the final output.
Why This Shift Matters for Everyday Writing
Punctuation may seem insignificant, but subtle stylistic choices shape how readers perceive tone, clarity, and professionalism. ChatGPT’s reliance on em dashes became a loud signal for anyone scanning text for AI footprints. With increasing pressure on authenticity, especially in digital publishing and corporate communication, users need tools that align with their personal writing voice.
This update is also a meaningful step in OpenAI’s broader strategy: giving people deeper control over how the AI behaves. ChatGPT already provides personality presets, memory features, and customizable writing preferences. The new punctuation rule strengthens the idea that AI should adapt to the user, not the other way around. It also hints at more advanced personalization features on the horizon.
Writers have struggled to get AI tools to consistently follow style rules. Many platforms, including Grammarly and Hemingway Editor, help users refine tone and structure, but generative AI tends to lean on ingrained patterns from training data. The em dash habit was one of those quirks. The ability to override it shows that these systems are finally becoming more customizable at a deeper behavioral level.
What This Says About AI Behavior and Model Design
The fix also highlights a bigger question within AI communities. If one punctuation preference took years to refine, how many other stylistic quirks still live inside large language models? Users often report that ChatGPT prefers certain sentence structures, transitional words, or formal tones. Some of these patterns come from the model’s training data. Others result from reinforcement learning and user feedback loops.
As OpenAI, Google, Anthropic, and other research groups continue developing generative models, personalization is becoming a priority. People want AI that sounds like them, reflects their communication style, and blends into their workflow without obvious AI fingerprints. This mirrors trends in enterprise tools like Notion AI and Microsoft Copilot, which focus on integrating AI into daily professional tasks.
What Users Can Expect Next
Now that ChatGPT can reliably drop its em dash habit, the next phase will likely involve more granular stylistic controls. Users may soon be able to choose preferences for sentence length, level of formality, emoji usage, vocabulary type, transition patterns, or even energy level in writing. Think of it like a personalized writing profile that stays consistent across emails, articles, documents, or creative projects.
This shift also encourages creators, students, and business leaders to explore how deeply they can mold AI to match their brand voice or personal tone. As AI generated content continues to flood the internet, authenticity and consistency will matter more than ever. Customizable language models can help people maintain a unique presence while still benefiting from AI assisted productivity.
The update also raises an interesting conversation inside the AI industry. If users can finally erase one of ChatGPT’s most recognizable tells, then the path toward true “invisible AI assistance” becomes clearer. The more the model adapts, the less it stands out inside professional or creative writing.
