Make Claude Write Like a Human With the Straight Talk Skill
What Arena found when it measured how Opus 5.5 writes against Opus 5, the one habit that got worse, and a Claude skill that turns every finding into a rule.
What Arena measured
Arena compared answers from Claude Opus 5.5 and Claude Opus 5 on Text Arena, at high reasoning level, from August to September 2026. It tracked 12 writing measures. 10 of them moved in a better direction: the new model uses far fewer of the tells that make text read as AI.
| Measure | Opus 5 | Opus 5.5 | Change |
|---|---|---|---|
| Em dashes (per 1,000 words) | 15.2 | 0.8 | 95% fewer |
| Semicolons (per 1,000 words) | 6.10 | 1.64 | 73% fewer |
| Words per sentence | 12.14 | 10.03 | 17% shorter |
| Long content words (share) | 41.7% | 38.6% | lowest of any Claude |
| Praise openers like "great question" (answers) | 3.76% | 2.60% | fewer |
| Agreement openers like "yes, exactly" (answers) | 3.76% | 2.16% | fewer |
| Honesty wording like "frankly" (per 1,000 words) | 0.48 | 0.31 | fewer |
| Stock phrases like "load-bearing" (per 1,000 words) | 0.19 | 0.11 | fewer |
| Abstract nouns (per 100 words) | 4.39 | 3.68 | fewer |
| Hedges like "perhaps", "arguably" (per 1,000 words) | 0.39 | 0.77 | 97% more |
| Answer length (words) | 453 | 481 | 6% longer |
Source:Arena, "How Claude Models Write", Text Arena outputs at high reasoning level, August to September 2026.
Keep reading
Free, for an email.
Unlocks 5 more sections, 2 copy-paste prompts, the document version, and every other guide on the site, for free. Enter your email once to keep reading.
19,000+ follow where these guides come from.No spam. Unsubscribe in one click.