One of the clearest improvements you can make to your AI outputs: whenever Claude has to guess at information, the output degrades. The fix is simple — give Claude examples.
1. Examples of what you want (positive examples)
"Here's an email I've sent that got a strong reply: [PASTE EMAIL]. Write 3 variations in a similar style for a different prospect."
Claude will match the tone, length, structure, and directness of your example far more accurately than any description of "write it like a peer, not a vendor."
2. Examples of what you don't want (negative examples)
"Here's an email I want to avoid sounding like: [PASTE GENERIC EXAMPLE]. Write a version that's more specific and less corporate."
Showing Claude what to avoid is sometimes more efficient than describing what you want, because the failure mode is specific and concrete.
Before running any high-stakes prompt (executive summary, CISO-level cold email, key objection response), find one example of something you've previously written or received that represents the quality you're aiming for. Paste it in. Ask Claude to match that quality for your new situation.
Abstract descriptions produce average outputs. Concrete examples produce outputs that match your standards.