A system prompt tells Claude to "write a good summary of each document." According to CCA-F prompt engineering principles, what is the primary problem with this instruction?

Question

Accepted Answer

B. "Good" is a subjective, unmeasurable criterion that produces inconsistent output and cannot be programmatically validated. Vague criteria like "good" are unmeasurable and lead to inconsistent, unpredictable output. Different invocations may produce summaries of different lengths, tones, and content focus — all technically compliant with "good." Explicit criteria (e.g., "3–5 sentences, include the main argument, key evidence, and conclusion, using formal tone") are measurable, automatable, and produce consistent results. Production systems require explicit criteria for reliable output.

A system prompt tells Claude to "write a good summary of each document." According to CCA-F prompt engineering principles, what is the primary problem with this instruction?

Related Questions

Discussion