What is Hallucination?
When an AI confidently states something false or made-up.
A hallucination is when a model produces an answer that sounds plausible but is factually wrong or fabricated.
It's why important AI output should be verified, and why grounding with retrieval (RAG) helps.
Hallucination is when an AI model produces information that is plausible-sounding but false — a fabricated statistic, a non-existent citation, a confidently wrong answer. It happens because language models generate text by predicting likely continuations, not by retrieving verified facts. When the model lacks the right information, it doesn't stop or say "I don't know" by default; it produces the most plausible-looking text, which can be entirely made up while reading as authoritative.
This is the single most important limitation to understand about generative AI, because the errors are not obviously wrong — they're fluent and confident, which is exactly what makes them dangerous. Hallucination can't be fully eliminated with current technology, but it can be reduced: grounding the model in real sources (retrieval-augmented generation), prompting it to cite or admit uncertainty, and using models tuned to be more honest all help. The durable defence, though, is human verification of anything that matters.
Why it matters
Understanding hallucination is essential to using AI safely. It's why you can't treat a fluent answer as a true one, why important outputs need checking, and why tools that cite sources or ground answers in your data are more trustworthy than open-ended chat. It reframes AI from an oracle into a fast, capable, but fallible assistant whose work you verify.
A concrete example
Asked for academic sources on a niche topic, a language model may produce a list of real-looking citations — plausible authors, titles and journals — that don't exist. Each looks authoritative, but the references were generated, not retrieved. A student who pastes them in without checking has cited hallucinated sources, which is exactly how this failure causes real problems.
Where you’ll meet it in AI tools
Hallucination is why research tools emphasise citations you can verify, why support-bot builders ground answers strictly in your documents, and why coding tools' output needs testing. When a tool advertises "grounded," "cited," or "sourced" answers, it's competing on reduced hallucination. Any tool that answers from open-ended generation alone carries more of this risk.
Hallucination: FAQ
Can AI hallucination be fixed?
Not fully with current technology, but it can be substantially reduced. Grounding the model in real sources (retrieval-augmented generation), prompting it to cite evidence or admit uncertainty, and using models specifically tuned to be more honest all lower the rate of fabrication. None of these make it zero, because hallucination stems from how generative models fundamentally work. So the practical answer is to reduce it with these techniques and verify anything important — treat AI as a fast assistant whose work you check, not an infallible source.
How do I avoid being misled by AI hallucinations?
Verify anything that carries consequences against a reliable source, and be most sceptical exactly where the model sounds most confident on specifics — names, numbers, citations, quotes — since those are common fabrication points. Prefer tools that show their sources so you can check, and notice that hallucination is likeliest when you ask about niche, recent, or precise factual details outside the model's strong areas. In short: use AI to draft and explore, but confirm facts yourself before relying on them.
Related terms
Find safety & ethics tools
Browse hand-reviewed AI tools — compare on pricing, features and real ratings.