Hallucination
A hallucination in AI is when a system, especially a chatbot, produces information that sounds confident and plausible but is actually false, made up, or not supported by any real source.
What is Hallucination?
One of the most important things to understand about today's AI chatbots is that they are not built to tell the truth — they are built to produce plausible-sounding language. A large language model generates its answers by predicting what words should come next, based on patterns it learned from huge amounts of text. Most of the time that produces useful, accurate responses — not because the text it trained on was carefully vetted (it's a messy mix of fact, fiction, and outright error), but because, across a vast amount of writing, true statements tend to recur and reinforce one another while any single falsehood gets contradicted or crowded out. So the most statistically likely continuation often happens to be the accurate one. But the model has no built-in sense of what is true and no way to check a fact against reality. When it doesn't actually know something, it doesn't reliably stop and say so — it often fills the gap with something that looks right. That confident, fluent, wrong output is what people mean by a hallucination.
The tricky part is that hallucinations don't look like errors. A traditional computer bug usually announces itself — the program crashes, or returns obvious gibberish. A hallucination arrives in the same calm, articulate, authoritative tone as a correct answer, which is exactly what makes it dangerous. The model will invent a statistic, a quotation, a historical date, a citation, or a product feature, and present it with no more hesitation than it shows for facts it has right. Some critics dislike the word "hallucination" itself, arguing it makes the model sound more mind-like than it is — the system isn't "seeing things," it's just generating plausible text that happens to be false — but the term has stuck because it captures how unsettling a confidently fabricated answer can feel.
Hallucination is one of the central unsolved problems in making AI trustworthy, and it is why you should never treat a chatbot's output as fact without checking anything that matters. The good news is that it can be reduced. Connecting a model to real source documents through retrieval-augmented generation, asking it to cite where its answers come from, keeping questions within what it reliably knows, and simply verifying important claims yourself all help. Newer models also hallucinate less often than earlier ones. But "less often" is not "never," and the responsible way to use these tools — particularly for anything involving health, money, law, or other people — is to treat the AI as a fast, fallible assistant whose work you still check, not an oracle.
Real-world example
A freelance writer is finishing an article and asks a chatbot for a few published studies that back up one of her points. In seconds it returns three references, each beautifully formatted — author names, journal titles, years, even page numbers. They look completely legitimate. But when she tries to look two of them up, they don't exist: no such paper, no such study, in some cases no such researcher. The model wasn't lying in any deliberate sense; asked for citations, it generated text shaped exactly like real citations, because that is what a confident answer to her question looks like. Had she pasted them straight into her article, she'd have published sources that were pure invention.
Related terms
Frequently asked questions
Why do AI chatbots hallucinate?
Because they're designed to produce plausible language, not verified facts. A chatbot answers by predicting likely next words from patterns in its training data, with no built-in way to check whether a claim is actually true. When it lacks solid information, it doesn't reliably admit the gap — it generates something that fits the shape of a good answer, which can be confidently wrong. Hallucination isn't a glitch bolted on by accident; it's a side effect of how these systems fundamentally work.
How can I tell if an AI is hallucinating?
You often can't from the answer alone — that's the core danger, since fabrications come in the same confident tone as correct answers. The practical defenses are external: verify any important claim against a trusted source, be extra skeptical of specific facts like statistics, quotes, dates, and citations, and watch for oddly precise details that would be hard to know. Tools that cite their sources help, because you can click through and check. The safest habit is to treat anything that matters as "unconfirmed until verified."
Are AI hallucinations getting better or worse?
Generally better, but not solved. Newer, more capable models tend to hallucinate less than earlier ones, and techniques like grounding answers in real documents and adding citations have cut the rate further. Even so, no current system is free of the problem, and more fluent models can sometimes make their mistakes harder to catch precisely because the wrong answers sound so reasonable. The sensible assumption is that hallucination is reduced but always still possible, so verification remains necessary.