01In plain English
A hallucination is when an AI model produces something that sounds right but is false: a statistic that does not exist, a quote nobody said, a court case or research paper that was never written.
02How it works
A language model writes by predicting plausible next words, and it has no built-in way to check facts. When it knows a topic well, the most plausible words are also true. When it is missing the answer, it still has to produce something, so it fills the gap with text that fits the pattern, such as a believable author name, a realistic-looking web address or a statistic that sounds right. It then writes it in the same confident tone as everything else. For example, you ask for a source on a niche subject, and it returns a perfectly formatted citation for a paper that does not exist. The model is not lying; it is completing a pattern.
03Why it matters when you are choosing
Hallucinations are why AI output needs checking before it is published or relied on. Tools that cite sources, or only answer from your own documents, make errors easier to catch.
04What to check
Check whether the tool shows its sources, whether you can limit it to your own material, and test it on questions you already know the answer to.
05Where you will meet it
06Common questions
Why does AI make things up?
A language model predicts likely wording rather than looking facts up. If it lacks the answer, it still produces something that sounds right. Rare topics, exact figures, recent events and requests for sources are the most common triggers, so those are the places to be most careful.
Does RAG stop hallucinations?
It reduces them but does not remove them. Giving the model your documents to answer from keeps it closer to real material, and lets it point to a source. It can still misread a passage or answer when the documents contain nothing relevant, so spot-check the citations.
How can I spot a hallucination?
Look for details that are very specific but unsourced, such as exact quotes, case numbers, links and statistics. Search for them independently. Ask the tool for its source, then open the source and confirm it says what the tool claims. If you cannot verify it, leave it out.