An opening time with practical consequences
Fictional example: a chatbot tells Jana that her local recycling centre stays open until 6 pm on Fridays. She plans to leave late in the afternoon. Jana checks that specific time on her municipality’s website, which she opens herself. In this invented example, winter hours differ from summer hours. The rest of the chatbot’s friendly explanation does not settle her decision. What matters is the correct centre, the relevant weekday and the season currently in effect.
Break a long answer into separate claims
Identify one statement that can be checked: who is supposed to be able to do what, and under which conditions? A paragraph can contain a correct explanation alongside a wrong number. Separate factual claims from suggestions and personal evaluations. An idea for organising a shopping list needs different scrutiny from a claimed application deadline. For important points, write down both the exact claim and what you would do because of it. Avoid assigning a single true-or-false verdict to an entire long response.
Go directly to the relevant source
For opening hours, consult the operator; for a product feature, current manufacturer documentation; for an official process, the responsible authority. Open the source yourself and read the relevant passage. A search preview or quotation without surrounding context is not enough for a decisive detail. Ask whether it covers the correct country, version, group of people and period. An authentic but older source may not fit your present situation. Look for limitations and footnotes as well as the prominent headline or main answer.
Follow-up questions aid research, not proof
You can ask the AI for a source or for areas of uncertainty. Treat the reply as a research lead. NIST describes how generative systems can produce plausible but invented statements and explanations. If the same chatbot changes its answer after a challenge, you still need to establish which version is correct. Two matching AI responses may repeat the same faulty information. Independent comparison remains essential; where the evidence conflicts, seek clarification from the responsible organisation or a suitably knowledgeable professional.
Recognise the boundary for consequential decisions
Personal professional assessment matters especially for health, legal deadlines or major money decisions. WHO highlights the risk of false or incomplete AI health information. An answer might help you prepare questions for a consultation, but should not lead you to start, stop or change treatment on your own. Do not upload confidential records merely to force a more certain answer. This page explains information checking and does not replace medical, legal or financial advice for your individual circumstances. Uncertainty can be an appropriate stopping point.
Your checklist
The checkboxes are for your own reference. No entries are sent to us. To keep a copy, print the page or save it as a PDF using your browser menu.
Common questions
Can an answer with references still be wrong?
Yes. It may misrepresent the source, use outdated information or cite something addressing another question. Open the reference and compare the specific claim.
Does asking “Are you sure?” make it reliable?
No. That may produce another wording, but it is not independent evidence. What matters is whether you can confirm the claim outside the conversation.
Do I need to check every harmless idea?
Match the effort to the possible consequence. Pay particular attention to numbers, dates, eligibility conditions and instructions on which you actually plan to rely.
Sources and further information
Primary sources from public authorities, consumer organisations and providers. A source may be in another language; follow the guidance that applies to your service and where you live.
- NIST: Generative Artificial Intelligence ProfileSource review date:
- CNIL: IA générative et vie privéeSource review date:
- WHO: Guidance on large multi-modal models for healthSource review date:
