An AI can give a confident explanation containing a made-up fact, citation, or detail. This is often called a hallucination. A polished answer can still need checking.
A believable answer is not a verified answer
Language models generate responses from learned patterns and the information available in the request. They do not automatically check every claim against a reliable source. Connecting search or documents can help, but does not remove every mistake.
Ask for something you can inspect
For a factual question, ask for a source and the specific passage supporting the claim. Open it yourself: a real link can still lead to a page that does not support the answer. For calculations, check the inputs and arithmetic.
Make uncertainty an acceptable answer
Try asking: “Use only these notes. If a detail is missing, say it is missing.” This makes the task clearer. It is still an instruction to test, not a guarantee that the model will obey.
Try this
Give the model a short paragraph with three facts. Ask four questions: three answered in the paragraph and one it cannot answer from the text. See whether it identifies the missing information rather than filling in the gap.
Use AI to help draft, organize, and investigate. Verify the claims that your next decision depends on.
Sources & further reading
These references go into more detail; some are written for developers.