10 Claude Mistakes Pro Users Are Still Making (And the Fix for Each)
Anthropic changed how Claude thinks. Most prompting habits from a year ago now cost you answer quality, and three of them cost it every single time. Here are all ten, with the fix for each, checked against the current API docs.
What changed, in one paragraph
Every current Claude model, Opus 5 and Sonnet 5 and the whole 4.6 family, now uses adaptive thinking. The model decides for itself when to think and how much, and on Opus 5 it is on by default: leave the thinking setting out of a request and it thinks anyway. The reasoning happens before the first word of the answer is written, and by default it is not shown to you. That one change is why a lot of the old prompting advice is not just useless now but actively harmful. The techniques below were written to make an older model plan, check itself and stick to a format. The new models already do those things, and the old scaffolding gets in the way.
Everything in this guide was verified against Anthropic's current model migration and prompt audit documentation on 2026-09-06. Where a claim is a judgement call rather than a documented fact, it is marked as one.
Keep reading
Free, for an email.
Unlocks 12 more sections, 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.