How to Check an AI Answer Before You Trust It

Start by finding the claims

Read the answer and highlight anything that could be true or false: names, numbers, dates, rules, definitions, quotes, prices, comparisons, and recommendations. These are claims. Claims need checking. Style suggestions and brainstorming ideas need judgment, but factual claims need evidence.

Ask where the answer came from

If the tool provides sources, open them and confirm that they actually support the claim. Do not trust a citation just because it exists. A source can be outdated, unrelated, misread, or too weak for the conclusion. If the tool does not provide sources, treat factual output as a lead to investigate.

Watch for current information

Some answers change quickly: prices, product features, laws, visa rules, tax rules, software interfaces, sports schedules, company leadership, and medical guidance. If the answer depends on time, verify it against a current primary source before using it.

Check numbers separately

AI can make arithmetic mistakes and can also use the wrong assumptions. Recalculate important numbers yourself. For tables, check totals, units, currencies, date ranges, percentages, and whether rows are missing. A clean-looking table can still be wrong.

Use a skeptical second prompt

After receiving an answer, ask: "What assumptions did you make? What could be wrong? What should I verify independently?" This does not prove the answer, but it often reveals weak areas. Then verify those areas outside the model.

How to apply this guide

Use this guide whenever an AI answer may influence a decision, publication, customer message, school submission, or business action. The more visible or costly the output is, the more explicit the checking process should be.

• Write the task in one sentence before opening an AI tool.
• Decide which parts need human review: check claims, verify sources, test assumptions.
• Remove private or unnecessary context before prompting.
• Check whether the final output changes a fact, promise, number, date, or decision.

Do not let a confident tone replace evidence. Separate useful wording from factual claims, then verify the claims through reliable sources or qualified review.

A safer prompt to try

Use this starter prompt when you want help with the idea in this guide but still want the model to show its limits.