Generate an Adversarial Review Process for Checking Your Own AI-Assisted Research

AI-generated research reads confidently whether or not it's right, and that confidence is the problem. This builds a repeatable checklist that hunts for contradicting sources, flags language covering a gap, and sorts every claim into verified, unverified and contradicted, ready to run on any answer you get next.

<context>
You are an AI output auditor who builds structured review processes for checking whether an AI-generated answer can actually be trusted, rather than accepting it at face value. The user regularly uses AI models for research on {RESEARCH_AREA} and wants a repeatable check they can run on any AI-generated answer.
</context>

<task>
**Generate** an adversarial review process:
1. Design a step where the user identifies every factual claim in the AI-generated answer that would need a source
2. Design a step for actively trying to find a source that contradicts each claim, not just one that confirms it
3. Design a step for spotting confident-sounding language covering an actual gap in the answer
4. Design a scoring or flagging system that sorts claims into verified, unverified, and contradicted
5. Give the whole process as a checklist the user can reuse on any AI answer in {RESEARCH_AREA}, not just this one
</task>

<output_format>
- The review process as a numbered, reusable checklist
- Scoring or flagging system explained
- One worked example applying the checklist to a short sample claim
- Length: checklist under 300 words, worked example under 150 words
</output_format>