AI Skill Report Card
Researching Thoroughly Before Answering
Quick Start11 / 15
Before answering, ask: "Would a wrong answer here be costly?" If yes, slow down:
- Break the question into sub-questions
- Gather information on each sub-question from multiple independent angles
- Cross-check facts against each other for contradictions
- Only answer once confidence is high and evidence converges
Recommendation▾
Examples are somewhat abstract—show actual research output snippets, source citations, or a worked sub-question decomposition rather than just describing what the output would contain
Workflow13 / 15
Progress:
- Step 1: Clarify the exact question being asked (avoid answering an adjacent question)
- Step 2: Decompose into sub-questions or claims that need verification
- Step 3: Research each sub-question thoroughly, using all available tools/sources
- Step 4: Identify conflicting information and resolve discrepancies
- Step 5: Cross-verify key facts against at least 2-3 independent sources
- Step 6: Synthesize findings into a coherent, complete answer
- Step 7: State confidence level and flag any remaining uncertainty
- Step 8: Deliver final answer with reasoning transparent enough to audit
Recommendation▾
Add a concrete bad-vs-good example pair (e.g., a rushed shallow answer vs. the thorough version) to illustrate contrast
Examples10 / 20
Example 1: Input: "What caused the 2008 financial crisis?" Output: A structured answer covering multiple contributing factors (subprime mortgages, securitization, regulatory gaps, ratings agencies, leverage) — each verified against multiple explanations, with interactions between causes explained, rather than a single oversimplified cause.
Example 2: Input: "Is this claim about a company's revenue accurate?" Output: Answer only given after checking the claim against multiple independent data points (e.g., financial filings, press releases, third-party reports), noting any discrepancies found and which source is most authoritative.
Recommendation▾
Include guidance on specific tools/techniques for cross-verification (e.g., how to search for disconfirming evidence, how to weigh source authority) rather than only general principles
Best Practices
- Prioritize accuracy and completeness over response speed — take as long as needed
- Actively search for disconfirming evidence, not just supporting evidence
- Distinguish between well-established facts, reasonable inferences, and speculation
- When sources conflict, explain the conflict rather than silently picking one
- State assumptions explicitly
- Give a final confidence assessment (e.g., "high confidence," "uncertain due to conflicting sources")
Common Pitfalls
- Don't stop at the first plausible-sounding answer — verify it
- Don't present speculation as fact
- Don't ignore contradictory evidence to preserve a clean narrative
- Don't sacrifice thoroughness for the sake of a quick reply
- Don't over-hedge to the point the answer becomes useless — commit to a conclusion once evidence supports it