Crushin’ Claude:
start with a useful prompt.
A short web companion to the downloadable guide. Read this here, then use the workspace to make the prompt your own.
1. Define the result
Start with the task, the audience, and what a successful answer must contain. “Help with this report” leaves too much unstated. Ask for the decision, summary, or structure you actually need.
2. Supply the right context
Give the model the relevant facts and constraints. Separate background material from instructions. For Claude, descriptive tags such as <context> and <instructions> can help keep those roles clear.
3. Make the format concrete
Specify the output structure and provide a short example when consistency matters. Include how the model should handle missing information; do not reward it for inventing details.
4. Check the result
Test the prompt on representative inputs, including difficult cases. Check important claims against their sources. When comparing variants, use the same inputs and a clear definition of success.
5. Save what works
Replace repeated details with variables such as {{audience}} and {{source_text}}. Save a template in the workspace and copy the completed prompt to your AI app. The workspace edits prompts; it does not execute them.