OpenAI quietly shipped a very human fix to a very nerdy problem: ChatGPT’s obsession with the em dash. What looked like a punctuation quirk became a cultural tell that something was “written by AI.” TechCrunch reports that, as of now, if you tell ChatGPT not to use em dashes in your custom instructions, it actually listens.
Sam Altman calls it a “small-but-happy win,” but it’s more than that. It’s OpenAI acknowledging that trust in AI isn’t only about accuracy or safety, it’s also about how the machine sounds when it speaks on our behalf.
For months, the em dash has been the AI equivalent of a Caller ID leak, an instant giveaway that ChatGPT was on the line. OpenAI has finally addressed this: users can now set “no em dashes” in custom instructions, and the model behaves accordingly. That seems cosmetic, but for anyone running brand-sensitive comms, legal docs, or customer-facing copy, this is a control knob that matters. It’s OpenAI responding to user irritation at the micro level, which in aggregate is what either builds or erodes confidence in using LLMs as silent ghostwriters across email, CX, and enterprise communications.
Market Impact
- LLM commoditization pressure: As models converge on raw capability, differentiation shifts toward fine-grained control over tone, style, and format. This update is a small but visible example.
- Brand voice & CX: Enterprises using LLMs in marketing, support, and CCaaS workflows gain more precise levers to enforce style guides. That reduces the “this sounds like AI” backlash from customers.
- Detection arms race: Simple stylistic heuristics (such as “too many em dashes”) become less reliable as detection signals, prompting regulators and platforms to shift toward watermarking, provenance, or behavioral analytics instead.
The Andy Perspective
What appears to be punctuation is actually trust infrastructure. If users can’t get the model to honor something as basic as “don’t use this character,” they won’t trust it on higher-stakes constraints. Fixing the em dash issue is a tell that OpenAI is prioritizing granular controllability—critical if LLMs are going to sit inside regulated, brand-sensitive communication channels.
What This Means To You
Audit any customer-facing or executive-facing flows where you rely on GPT-style models for drafting:
- Add explicit tone/punctuation constraints into your system or custom instructions (e.g., “no em dashes, short sentences, AP style”).
- Run A/B comparisons on existing templates (email, support macros, briefings) to ensure the model now respects those constraints.
- Update internal “prompt playbooks” for staff and vendors to incorporate these new, more reliable style controls.
GTM Angle
If you are selling AI-enhanced communications, CX, or CCaaS:
- Position this as “brand-safe AI copy”: emphasize your ability to enforce client style guides down to punctuation.
- Offer a “voice hygiene” audit to prospects—scan their outbound comms, then show before/after using strict style constraints powered by LLMs.
- For partners wary of “AI tone leakage,” this is a proof point that you can tune the machine to speak like them, not like the generic internet.