What is an AI hallucination and how often does it actually happen?
An AI hallucination is when a language model confidently generates information that is completely made up, factually incorrect, or nonsensical, presenting it as if it were true. Think of it less like a lie and more like a very convincing improv actor who never breaks character, even when they have no idea what they're talking about. It happens because the AI's core job is to create a plausible-sounding sequence of words, not to retrieve verified facts from a database. It's a storyteller first, a researcher second. How often it happens is tough to pin down with one number. A 2023 study by Vectara found hallucination rates ranging from about 3% to 27% depending on the specific task and model. In my experience, you'll see it most when you ask for very specific dataālike a detailed financial figure from a 2019 quarterly reportāor a biography of a semi-obscure person. The model will just invent a plausible-sounding number or life event. A classic example is asking for a list of books by a famous author and getting back a title that doesn't exist, complete with a convincing summary. The most practical tip I can give is to treat the AI's output as a first draft that requires fact-checking, especially for anything with numbers, dates, or direct quotes. **Related**: How do I stop AI from making up facts? | Can I trust AI for research and homework?