A hallucination is when an AI model states something false or made up as if it were true. It might invent a fact, a quote, a citation or a number, and it says it with the same confidence as everything else. Picture asking a chatbot for sources to support a grant application and getting back a neat list of reports that sound real but do not exist. This happens because language models are built to produce text that sounds plausible, not to check whether it is correct.
Why it matters
Hallucinations are the main reason a person needs to read and check AI output before it goes out or anyone relies on it. For a small organization, the risk shows up in donor letters, grant reports, policy summaries and anything with names, figures or references in it. Treat every specific claim as a draft until someone has confirmed it. When a vendor offers you an AI tool, ask how it cuts down on made-up answers, for example whether it uses grounding or retrieval-augmented generation, and whether it shows where each answer came from. Also watch for overreliance on your own team, where people stop checking because the answers usually look right.
Published Updated