Glossary
Hallucination
A hallucination is when an AI states something false with full confidence: an invented fact, source, law, or number. It happens because AI predicts plausible text, not verified truth, so important facts always need checking.
Ask an AI for the opening hours of a small-town pharmacy and you may get a precise, polite, completely invented answer. That’s a hallucination: not a glitch, but a direct consequence of how an LLM works. It produces the most plausible next words, and a confident wrong answer is often more plausible-sounding than “I don’t know.”
The dangerous part is the tone. Hallucinations arrive in the same calm, fluent voice as correct answers, complete with fake article numbers and fake URLs. The model isn’t lying; it has no notion of true and false to lie about.
Honest ground rules: modern models hallucinate less, and web-search grounding and RAG help a lot, but nobody has eliminated it. So: brainstorm, draft, and summarize freely; verify names, numbers, laws, prices, and citations before you act on them. Asking the AI to cite sources you can click is the single best habit.
Where you’ll meet this
Anywhere an AI sounds certain about checkable facts. ChatGPT, Claude, Gemini all do it. Prompts that say “if unsure, say so” measurably help; our prompt fixer builds that guardrail in.