How do I fact-check AI answers and avoid hallucinations?

AI models are trained to sound confident even when guessing, so a fluent answer with citations is not proof: open the sources and check they exist and actually say what the AI claims. Use lateral reading (verify names, numbers and dates in an independent search), ask the model for counter-evidence, and scale your rigour to the stakes; anything going to investors, customers or a court gets line-by-line verification. Grounded tools (Perplexity, deep research modes, NotebookLM) hallucinate less than bare chatbots, but nothing removes the need to check.

A quick orientation. The real value is below: resources worth your time, from people who've actually done it.