RIP em dashes: ChatGPT just made AI writing harder to spot
Writers have had the same gripe for years: ask ChatGPT for clean copy, get peppered with em dashes that read like dramatic pauses. That quirk became a tell, especially in client work and email. OpenAI just gave us a simple fix that actually sticks.
Highlights
- User-level control to disable ChatGPT's overused em dashes
- A clear step toward deeper style tuning and personalization
- A small punctuation change that makes AI output feel more human
Here is what changed. If you tell ChatGPT to avoid em dashes in Custom Instructions, it now follows that request with much better consistency. The update rolled out on November 15, and it pairs neatly with memory and other personalization hooks, signaling a shift away from one-size-fits-all prose.
Sam Altman called it a "small-but-happy win." That fits. The character caused enough editing pain to deserve a proper send-off.
Why this matters for writers
- Fewer tells in pitches, drafts, and client emails.
- Cleaner rhythm that matches brand voice and publication style.
- Less time sanding down awkward pivots and pauses.
How to switch off em dashes in ChatGPT
- Open Settings in ChatGPT on the web browser or desktop app.
- Select Personalization.
- Ensure Enable customization is on.
- In Custom Instructions, add a note like: "Please do not use em dashes in any response."
- Save, then start a new chat so the preference applies cleanly.
Working across devices? Repeat the same idea on mobile under Settings > Customize ChatGPT. The preference ties to your account, but a fresh thread still helps.
If you want more detail, OpenAI's guide covers Custom Instructions here: official instructions.
What still trips readers and detectors
Em dashes were a big tell, but far from the only one. Stock phrases, over-explaining, rigid structure, and tidy concluding lines can flag machine-written text fast.
- Ask for shorter sentences and fewer hedges.
- Set a reading level, tone, and audience for each project.
- Define punctuation rules: serial comma on, semicolons rare, no exclamation points.
- Request varied sentence openings and minimal filler transitions.
A quick Custom Instructions template
- Style: Avoid em dashes. Prefer commas or parentheses. Keep sentences under ~22 words on average. Use active voice. Cut filler like "moreover" and "furthermore."
- Punctuation: No exclamation points. Use the serial comma. Limit semicolons. Single spacing after periods.
- Voice: Concise, conversational, specific. Mirror my prompt's tone unless stated otherwise.
Why the overuse happened in the first place
OpenAI has not unpacked the technical reason, which is fair. The likely causes are training data and reinforcement patterns that rewarded dramatic pauses in general-purpose writing.
Backstory is nice, but the practical win is better. You now have an off switch.
Use the em dash like a pro
The em dash is not canceled. It just goes back to where it belongs, used sparingly when rhythm calls for a punchy aside. If clarity is the goal, commas or short sentences usually do more with less.
- Need a refresher on the mark itself? Em dash overview
- Want to sharpen your ChatGPT workflow for writing? Complete AI Training: ChatGPT
Simple change, real impact. Set your rules once, then write in your voice without babysitting punctuation every paragraph.
Your membership also unlocks:
