FIX · ALL AI TOOLS
AI made something up: how to catch it
A chatbot doesn't look facts up; it predicts the most likely next words. So sometimes it produces a convincing book title, a link that goes nowhere, or a wrong number, all in the same confident tone.
In 30 seconds
- Every AI does this sometimes. You didn't use it wrong.
- The more specific and obscure the detail (names, dates, amounts, laws, URLs), the more likely it's invented.
Get fewer made-up answers: add this line
Only answer with information you're sure of. If you're not sure, say "not sure". Don't guess. List your sources.
- Give it the material. To summarize an article or contract, paste the text in instead of asking from memory.
- Need sources? Use a tool that shows them, like Perplexity, and open at least one.
Check these yourself
- Numbers, dates and amounts against the original.
- Names of people, books and organizations: search that they exist.
- Links: open them. Delete any that don't work.
- Legal, medical and money questions: use AI for background, then ask a professional.
Rule of thumb
- The more confident it sounds, the more worth checking it is.
- Never send unchecked AI facts to a boss, a client or the internet.