1. Why convincing is not the same as correct

Generative AI is trained to produce plausible responses, not to guarantee truth. It can confuse two people, cite a report that doesn’t exist, use an outdated price or make an arithmetic error. Its professional tone makes those mistakes harder to spot.

For a recipe suggestion, a rough idea may be fine. For medication, workplace safety, legal duties, investments or travel requirements, you need stronger evidence. The level of checking should rise with the consequences of being wrong.

2. The five-step verification habit

1. Mark the claims: highlight names, dates, numbers, rules, quotations and anything you plan to rely on.
2. Ask where they came from: request original sources, not just a confident restatement.
3. Open the sources: a link or citation isn’t proof unless the page exists and actually supports the claim.
4. Cross-check: for important facts, compare with an authoritative, up-to-date source and check the date or location.
5. Decide: keep what is supported, correct errors and mark what remains uncertain.

3. A simple example

Suppose you ask, “What is the current train fare?” The answer could depend on the date, route, ticket type, operator and live availability. Even an AI with web access may misunderstand the exact journey. Use its answer as a starting point, then check the booking service before paying.

Another example: an AI says a certain UK workplace law requires action within a precise number of days. Because a wrong deadline can cause harm, you should verify it directly against the current regulator or official guidance, not trust the chat alone.

4. Ask better research questions

This improves the task, but it doesn’t replace your review. Some assistants may only provide sources when a browsing or research tool is actually enabled; others may output links that don’t support the text.

5. A quick risk test

Ask yourself: What’s the worst that happens if this is wrong? If the answer is embarrassment or a slightly odd dinner, light checking might be enough. If it affects someone’s wellbeing, money, rights or safety, use official sources and qualified people.

Try it yourself

Choose a public, checkable fact: for example, the opening hours of a museum or the launch year of a household invention. Ask an AI for the answer and a source. Then open the source yourself.

Record whether the source existed, whether it supported the claim and whether the answer needed correcting. Do not use personal or confidential information.

Can you explain it back?

Answer each question before opening the explanation. You aren’t being graded — this is how you make the knowledge stick.