Error Recovery Patterns

Handle AI errors, hallucinations, and unexpected outputs with resilient prompt design.

Category: Engineering | Type: Skills

Skills: Error Handling, Hallucination Detection, Resilience

Techniques: Structured Output, Constraint-Based, Self-Verification

Prompt

AI outputs will fail. Build resilience in: 1. Verification Prompts — after generation, ask the model to fact-check its own output on [your topic/domain]. 2. Confidence Scoring — append "Rate your confidence 1-10 and explain why" to detect uncertainty. 3. Contradiction Detection — ask the model to list ways its answer could be wrong. 4. Source Attribution — require citations for every claim. No source = flagged. 5. Fallback Chains — if the first prompt fails format validation, retry with a simplified version. 6. Adversarial Testing — regularly try inputs designed to produce errors. 7. Output Validation — parse, type-check, and constraint-check every response before using it. Treat AI output like user input: never trust, always verify.

Browse all prompts at Ask Wisely