How to make Claude write human-facing documentation?
retained text
Hey community, With the new model generation, starting with Opus 5, the output quality considering the AI-human interaction has changed. This is not only noticeable during the conversation, but as well for documentation. **Opus 5** is mentioned, since the output prompts were known to be truly challenging. Although other models are better in this regard, they are also following comparable patterns. Most noticable, are for me: \- the wall of text \- no readers guidance \- no logical flow considering the information \- information hidden im the text \- lazy usage of the markdown style \- sections with no clear ownership, \- list which should have been subsections \- models are now preventing \`—\` but using \`;\` instead \- \`;\` is then used for connecting statements or other listings in a hard to understand format \- no progressive disclosure \- conversation context spillover **Is there anything I'm missing out on how to make Claude write human facing documentation?** For some time now, I have tried to fix this with all kinds of ideas, skills, rules, output styles, and what have you, but not truly with success. Success means: Claude writes nice documents, like \`README.md\`, \`ROADMAP.md\` , ADR.md\`, or even Jupyter Notebooks \`\*.ipynb\` , with out detailed instructions and updates them accordingly. Aside from the commonly known resources, like: \- ASD-STE 100, \- Diátaxis \- Googles technical writing for developers \- etc. I've tried all kinds of skills, output styles, and others; none of them led to real success. And to be honest, I'm so sick of running canary runs with no noticeable improvement.
document fetched 2026-11-03 00:41 · canonical https://reddit.com/r/ClaudeCode/comments/1xm0kl8/how_to_make_claude_write_humanfacing_documentation