How to recognize AI hallucinations
AI systems often sound confident in their answers, even when the answer is made up. Here's how to spot and avoid that.
A "hallucination" is the term for when an AI system, like ChatGPT or Claude, states something that sounds convincing but is factually wrong or made up — citing a book that doesn't exist, giving the wrong date, or inventing a source that isn't real. The system doesn't "know" it's wrong; it simply calculates which answer sounds plausible, based on patterns from its training.
This happens more often with specific, checkable facts (names, dates, numbers, sources) than with general explanations. The more specific and obscure the question, the more likely the model is to fill in a gap instead of honestly saying "I don't know."
A few simple ways to spot and avoid hallucinations: ask the system to name a source and check that source yourself. Be extra critical of numbers, statistics, names, and quotes — exactly the kind of details that sound convincing but are easy to fabricate. If needed, ask the same question a different way and see if the answer stays consistent. And for anything with real consequences — medical, legal, financial — use AI as a starting point, not a final judgment.
Hallucinations aren't an occasional bug that will disappear soon; they're a fundamental feature of how language models work (see our article "What is a language model?"). The best remedy isn't blind trust, but healthy skepticism — especially for the kind of facts you'd normally look up yourself anyway.