AI Overuses the Em Dash? Analysis of 88,927 Chats Confirms It
Pristine_Ad3669 · reddit · 2026-08-30
An analysis of 88,927 AI chats confirms the common observation that LLMs overuse the em dash (—):
- Usage Gap: Only 2% of user prompts contain an em dash, compared to 34% of LLM responses.
- Historical Context: Google Ngram data shows em dash usage in U.S. books peaked in 1979 and declined to 1800s levels by 2006-2022.
- Search Trends: Google search interest for the symbol grew 245% in the past year (Aug 2025-Jul 2026) and 44% in the prior year, coinciding with AI adoption.
Why it happens:
- Token Efficiency: It can sometimes connect words using one fewer token than alternatives.
- Perceived Sophistication: It is associated with polished, academic, or complex prose.
- Human Preference: Trainers and annotators who shape model output favor it, so the model learns to reproduce it.
More from Models
- OpenAI stealth-testing mysterious internal model 'Aeon' in Codex desktop app and Dots — cedric_chee · 2026-10-16
- Argon edges out astra as daily driver, but Claude still wins heavy coding work — iruletheworldmo · 2026-10-16
- NVIDIA ships FP4-quantized Qwen3.8-Flash for local use, trending on Hugging Face — nvidia · 2026-10-16
- RoboVista benchmark: Gemini-4 Argon hits 81.9% on robotics understanding — berkeley_ai · 2026-10-16
- Dev reports decision model jev beats gpt-6-luna on quality in real training pipeline — genmon · 2026-10-16
- Google won't publish Argon 4 limits in AI Pro—users compare experiences — Consistent-Text-2662 · 2026-10-16