Fix AI prompts that produce generic output

Diagnose and fix AI prompts producing generic or wrong output

When AI output is disappointing and the fix is in the prompt, not the model — saves time vs trying new models.

• · current prompt, actual output, desired output specification

• · fixed prompt with 10 diagnostic patterns addressed, annotated changes

• · LLM (Claude/Perplexity/web-search)

User can articulate what they wanted vs what they got; original prompt available

• · Fixed prompt still produces generic output (context insufficient)
• · Hallucinated facts persist (model issue, not prompt issue)
• · User conflates negative instructions with positive (harder to follow)

• · 10-pattern diagnostic checklist (generic, hallucination, misunderstanding, length, instruction-ignoring, inconsistency, non-actionable, non-critical, multi-task, reuse fragility)
• · Distinguishes negative vs positive instructions
• · Enforces evidence/citations in output format