stale · formatting · first recorded 2026-07-02 · registers: hn, cooking, parenting
Models emit “curly” quotes where web-form typists produce "straight" ones — which looks like a tell until you notice the confound: iPhones, Word, Google Docs and every CMS smarten quotes automatically. Our canon of published human prose is full of them. Flagging curly quotes flags devices and editors, not writers — so we don't.
Kept public as a caution, like the em dash: production-pipeline artifacts make terrible evidence.
⚠ also common in good human writing (collateral score 0.53 — the closer to 1, the more it also fires on acclaimed pre-AI prose)
…one of those rare “boring is good” moments…
occurrences per 10k words, measured 2026-09-28 (run #15, 3 current + 3 vintage models)
| current AI | vintage AI (2024/25) | humans (matched) | canon (great pre-AI prose) | |
|---|---|---|---|---|
| pooled | 95.57 | 10.18 | 6.01 | 50.93 |
| hn | 111.29 | — | 10.80 | |
| cooking | 57.30 | — | 3.36 | |
| parenting | 144.28 | — | 0.40 | |
| 32.67 | — | — |
| run | date | AI /10k | vs humans |
|---|---|---|---|
| #2 | 2026-07-02 | 76.40 | ×18 vs humans |
| #3 | 2026-07-02 | 91.71 | ×14 vs humans |
| #4 | 2026-07-06 | 74.18 | ×19 vs humans |
| #5 | 2026-07-13 | 70.35 | ×7.8 vs humans |
| #6 | 2026-07-20 | 90.51 | ×24 vs humans |
| #7 | 2026-07-27 | 90.18 | ×15 vs humans |
| #8 | 2026-08-03 | 88.90 | ×14 vs humans |
| #9 | 2026-08-10 | 96.07 | ×15 vs humans |
| #10 | 2026-08-17 | 85.86 | ×15 vs humans |
| #11 | 2026-08-31 | 89.58 | ×15 vs humans |
| #12 | 2026-09-07 | 87.83 | ×11 vs humans |
| #13 | 2026-09-14 | 92.40 | ×34 vs humans |
| #14 | 2026-09-21 | 90.40 | ×15 vs humans |
| #15 | 2026-09-28 | 95.57 | ×16 vs humans |
regex: [“”]