Beginner · Level 1: fundamentals
How to check an answer from AI
An answer can sound sure and still be wrong. Learn what to always check, how to ask for sources and why you open them yourself instead of just reading them.
What you will learn
- Know which kinds of information need checking every time
- Ask for sources and open them to see whether they say what is claimed
- Involve a human for medical, legal and financial topics
In the first lesson you saw that a chatbot produces plausible text, not verified text. The practical conclusion: you are the last filter. The good news is that checking does not take long if you know what to look for. This lesson gives you a simple habit in four steps.
What you always check
- Figures and calculations: amounts, percentages, deadlines, doses.
- Dates, names, titles and addresses: people, institutions, books, articles, links.
- Quotes: if “someone said” something, look for the sentence at the source.
- Rules and laws: they change, and a model can cite an old or invented version.
Step 1: ask for sources
You can ask directly: “For every claim with figures or dates, give me the source and the link.” A service that has a search tool will show real pages. One that does not may invent links that look fine and lead nowhere. That is why step 2 follows.
Step 2: open the source
A link is not proof until you open it. Check three things: the page exists, it belongs to a source you trust (an institution, a publication, an official site) and it truly says what the chatbot claims. Sometimes the page is real but talks about something else. It also happens that a domain that responds belongs to someone other than you expect.
Step 3: ask it differently
Ask the same question in a different way, or in a new conversation, and compare. If the answers differ on facts, do not rely on either. You can also ask: “What in your answer is the least certain? What should I check?” Sometimes it will point to weak spots, but that is no guarantee.
A prompt for checking: “Above you wrote some claims with figures and dates. Make me a list of each claim, then for each one say how sure you are (sure, probably, not sure) and which official source I should open to check it. If you are not sure about a source, say so directly; do not invent a link.”
Step 4: put a human in the loop
For decisions with consequences, reading one source is not enough. For medical, legal, tax and financial topics, use AI only to prepare yourself: to understand the terms and learn which questions to ask. Then you talk to a doctor, a lawyer, an accountant or the competent institution.
The lessons on this site are educational and do not replace a specialist’s advice. Do not delay a doctor’s visit or a legal decision because a chatbot gave you a reassuring answer.
If you publish a text written with AI, you are responsible for what appears. In the European Union, the official AI Act page notes that some AI-generated content, for example realistic fakes (deepfakes), must be clearly labelled. Rules change, so check the official page before relying on them.
Key takeaways
- A sure tone proves nothing; check figures, dates, names, quotes and laws.
- Ask for sources, then open them yourself and see whether they say what is claimed.
- Compare answers to differently worded questions and ask for the uncertain points.
- For health, law, taxes and money, AI prepares you, but a specialist decides.
Try it now
Replace what is in braces with your own details, then copy.
Check what you learned
Three questions. They count for nothing but you.