Everybody thinks they can spot AI, and when it comes to text, the favorite signal is em dash usage. I was one of those people.
I put this report together because I recently gave a recommendation on removing em dashes to sound more like a human, I wanted to make sure I wasn’t full of it, but it turns out I was š¬.
I analyzed em dash usage on the daily top 10 articles in Hacker News since 2018 (80M words). Here are the details.
What I expected:
A jump after November 2022 when ChatGPT launched. Maybe 20-30% higher em dash usage as AI-assisted content flooded the web, then a drop after people started declaring that em dashes were dead in 2024. The reality was the exact opposite.
What I found:
- 2019: 1.40 (peak)
- 2022: 0.96 (ChatGPT launches in November)
- 2023: 0.93 (-3%)
- 2024: 0.82 (lowest in the dataset)
- 2025: 1.21 (+48%)
- 2026: 1.27 (partial year)
Dashboard link (methodology included)
Interpretation:
2024’s drop might reflect writers consciously avoiding em dashes after reading all those “how to spot AI” articles. The 2025 spike could be increased AI usage in content creation, or it could be natural variation returning to historical norms.
Either way, the “em dash = AI” heuristic is wrong. 2019 had the highest em dash usage in my dataset, three years before ChatGPT existed.
Leave a Reply