Spotting mistakes and made-up facts
Why AI "hallucinates" and a quick routine to catch it.
When a chatbot states something false as if it were true, people call it a hallucination. It isn't lying; it's producing text that sounds right.
Where mistakes hide
- Numbers and dates: prices, statistics, years.
- Names and quotes: who said what, book titles, people's job titles.
- Sources and links: references that look real but don't exist.
- Recent events: anything after the tool's training, unless it searched the web.
- Maths: long calculations can go wrong.
The 30-second check
- Is this important? If you'll send it, publish it, pay for it or decide something with it: check.
- Ask for the source: "Where does this come from? Give me a link." Then open the link.
- Ask it to double-check: "Check your answer for mistakes." It often catches its own errors.
- Cross-check one key fact with a search engine or an official site.
Tools that show their sources
Some tools are built to cite: Perplexity shows web sources next to each answer, and NotebookLM answers only from documents you upload, with clickable citations. Use them when facts matter.
Red flags
- Very specific numbers with no source.
- A quote that's perfectly on-topic.
- An answer that changes when you ask again.
None of this means "don't use AI". It means: use it for drafts and thinking, and check the facts you rely on.
Key takeaways
- Numbers, names, quotes, sources and recent events are where mistakes hide.
- For anything important: ask for the source and cross-check one fact.
- Use citing tools like Perplexity or NotebookLM when facts matter.
Try it now
Ask a chatbot: "Give me three facts about the history of [your city], each with a source link." Open every link.
Open PerplexityFree · paid from $20/moYou've got it when: You checked every link and know which facts are supported and which aren't.