AI can make a weak answer look finished. Good formatting, confident language and lots of detail are not proof that the output is correct.
Before you send, publish, buy, build from or act on AI-generated work, review the parts that can actually fail.
Not sure how much checking this particular output deserves? Use the free AI Output Review Router. It routes the answer to a proportionate review level based on evidence, freshness, external-action claims, consequence and reversibility.
What should you check in an AI answer?
- Objective: did it answer the task you actually gave it?
- Facts: can important factual claims be supported or independently checked?
- Assumptions: has a guess quietly become a fact?
- Missing context: could omitted information materially change the answer?
- Source fit: if files or sources were used, does the answer reflect what they actually say?
- Freshness: does anything depend on information that may have changed?
- Capability: is the AI claiming it accessed, changed or verified something it could not actually do?
- Consequence: does the amount of verification match the cost of being wrong?
- Completeness: are required constraints, deliverables and edge cases present?
- Actionability: is the next step clear?
The most important distinction: review is not verification
Reading an answer twice does not verify it. If a claim matters, open the source, check the number, confirm the date or compare the output with the authoritative material.
The higher the consequence of an error, the stronger that verification should be. A brainstorming list and a decision involving money, safety or professional reliance should not pass through the same review gate.
Four useful outcomes
- PASS: reviewed and fit for the intended low or normal-risk use.
- REVISE: useful, but material issues need fixing.
- VERIFY: evidence or external checking is still required.
- REJECT: fundamentally wrong, misleading or not fit for the intended use.
A checklist cannot guarantee correctness
A quality gate improves review discipline; it does not turn uncertain output into truth. High-consequence legal, medical, financial, safety, security or specialist decisions need appropriate authoritative sources and qualified review.
Choose the review strength first
Run the free AI Output Review Router if you want a quick decision between light review, source verification, external-action verification and a stronger stop-and-verify route.
Want the complete reusable gate?
The AI Output Quality Checklist is £7 and contains the complete 12-point quality-control gate, a 60-second human check, copy/paste AI-assisted auditor, examples, quick start and full guide.
Before you trust the output, test the output.
