Why do AI chatbots sometimes make up fake facts or citations?

AI chatbots make up fake facts because they don't 'know' things the way you and I do. They're prediction engines, not databases. When you ask a question, the model isn't searching a library of facts. It's predicting the next most likely word in a sequence, based on the patterns it learned from its training data. This quirk is often called 'hallucination,' though I find that term a bit too mystical. It's really just a confident-sounding guess that turned out to be wrong. For example, if you ask for a biography of a minor historical figure, the model might generate a perfectly formatted, plausible-sounding life story with a birth date, a hometown, and a career path. It feels real because the pattern is correct, but every single detail could be fabricated. The model saw millions of real biographies during training, so it knows a biography should start with 'was born in...' and mention a profession. It just doesn't have the actual data for that specific person, so it fills in the blanks with statistically likely words. I've seen it invent court cases, academic papers with fake DOIs, and even quotes from books that don't exist. A good mental model is to think of an LLM as an improv actor who has read the entire internet. The actor is brilliant at staying in character and sounding authoritative, but they will never break character to say, 'I actually don't know the answer to that.' This is why you should always double-check any factual claim an AI makes, especially if you plan to publish it. Treat its output as a very fast first draft from a confident intern, not a finished, verified document.