Using Claude to Develop Your Prompts

Here's a technique most people miss: before executing your research prompt, ask Claude to critique it first. Send your draft prompt and say "Before you execute this, tell me what's ambiguous, what's missing, and what would make this produce better results."

Claude will identify gaps in your brief — missing scope boundaries, unclear success criteria, format ambiguity. Fix those, then execute. You get better results in fewer iterations because the prompt itself was refined before any research happened.

I'm about to ask you to do competitive research. Before you
execute, critique my prompt:

[paste your research brief here]

Tell me:
1. What's ambiguous that could lead to wrong results?
2. What scope boundaries am I missing?
3. What would you add to get higher-quality output?
4. Is the requested format the best choice for this data?

This creates a critique loop: Draft prompt → Claude critiques → You refine → Execute refined prompt. The critique step typically catches 2-3 issues that would have required follow-up questions after seeing the first output.

You can also use this pattern recursively. After getting research results, ask Claude to critique its own output: "What's the weakest part of this analysis? Where are you least confident? What would a skeptic challenge?"