Technical writing
Build docs around real reader tasks, write for a named reader, define jargon, and open with an example that runs.
-
How Do You Get Claude to Explain the Jargon in Your Docs? Tell Claude who the reader is and what they already know, then ask it to find every term that reader might not understand and suggest a short definition at first use. Level 2: Contextreader may not know it →Draft it, then verifycheck with the term's ownerDefine from your glossaryyour meaning, at first useLeave it alonethe reader already knows itUse it consistentlyno definition neededin your glossary →
-
How Do You Get Claude to Explain Something at the Right Level? Describe the actual reader: their job, what they already know, and what they'll do after reading. Level 2: Context
- Executivewhy it matters
- Practitionerhow to use it
- Implementerhow it works
-
How Do You Get Claude to Write Docs With Examples That Run? Ask Claude to open the page with the smallest example that actually works, then explain it, then state the rule. Level 2: Context
- 1Smallest working exampletop of the page
- 2What it shows
- 3The general rule
- 4Edge cases
-
How Do You Use Claude to Write Docs People Actually Use? Give Claude evidence of what readers are really trying to do, such as support tickets, search terms or onboarding questions, and ask it to organize the docs around those tasks. Level 2: ContextFeature-first docsfollows the interface
- Lists every option
- Reader assembles the steps
- Doesn't say what goes wrong
Task-first docsfollows the reader's goal- Opens with what they want done
- One path from start to finish
- Names the common failure
Give it the real material
A technical doc has to match the product, and Claude can't know your product unless you show it. Attach the reference, source file or changelog, and tell Claude to use only that for facts. Anthropic's guide to reducing hallucinations suggests having Claude find a supporting quote for every claim, and remove what it can't support. For code, ask it to run the example where it can, and to tell you when it can't. Its sandbox has limited network access, which varies by plan.
Start from the reader
The cards here all begin with a named reader and their task. Say who's reading and what they'll do next. Use support tickets or search terms to find out what they need, lead with an example that works, and define terms at first use.
Can Claude write documentation from my code?
Yes, if you give it the code. Paste the source or upload the file and describe the reader and their task. Tell it not to use any function or parameter that isn't in what you gave it, then run the examples yourself before publishing.
Does Claude test the code examples it writes?
Only if it actually runs them. With Code execution and file creation on, it can run code in a sandbox, but private packages and your internal systems usually aren't reachable from there. Treat a code block with no shown output as untested.
How do I get Claude to write for non-technical readers?
Describe the reader's job, what they already know and what they'll do after reading, rather than saying keep it simple. Ask for terms to be defined at first use. In Claude for Word, you can select one section and ask to simplify it for a non-technical audience.
Can Claude keep terminology consistent across a document?
Ask it to check. Anthropic's Word page gives prompts such as flagging inconsistent defined terms, and asking whether the same party is called different names. Keep a glossary in project knowledge, so that Claude uses your definitions.