How much trust does this answer deserve?
Start with three questions
You do not need a technical checklist. These three questions catch many of the ways an AI answer can be polished but still not ready to use.
Does it fit?
Did it answer the real question and respect the person, task, constraints, geography, timeframe, and context?
What supports it?
Which parts come from the source or known facts? Which parts are inference, assumptions, suggestions, or unknowns?
What if it is wrong?
A weak brainstorm is different from an error that affects eligibility, funding, rights, a board decision, or someone’s career.
Try the same judgment skill in five different situations
Choose a scenario, read the AI response, then decide what kind of trust it deserves. The point is not the label—it is noticing why.
You can ask AI to show its work more clearly
When an answer feels too smooth, vague, confident, or convenient, try one of these moves.
Useful when the answer seems to know more than the source.
Useful when evidence and interpretation are blended together.
Useful when only part of the response may be risky.
Useful for policy, eligibility, funding, data definitions, or other governed work.
Useful when an answer sounds too certain.
Useful for board briefs, data stories, employer outreach, and career advice.
Not every answer needs the same level of verification
Brainstorming, formatting, agendas, first drafts. Review for usefulness and obvious errors.
Career options, employer communications, data interpretations, board briefings, training comparisons. Verify important facts, context, and assumptions.
Eligibility, service denial, benefits, accommodations, sanctions, legal interpretation, or other decisions affecting rights or access. Use controlling sources and the required human decision process.
You do not need AI to be perfect before it can be useful. You need to know what it is doing, what the answer rests on, and how carefully you should check it before you rely on it.