Fake citations: why AI answers with references can still be wrong

When an AI generates an answer with citations, it may look credible, but the citations themselves can be completely fabricated. For example, the model might invent a journal article title, author, and year that sound plausible but do not exist. Alternatively, it might cite a real source but twist what it actually says, like claiming a study supports a conclusion when the study says the opposite. This is different from a simple factual error—it is a hallucination that creates false evidence, making the answer seem more trustworthy than it is.

To spot fake citations, check if the source is real by searching the title and author. If you cannot find it, treat the claim as unsupported. Also, look for vague or generic references like 'a 2023 study found…' without specifics—this is a red flag. In an exam context, if a question asks about verifying AI outputs, remember that you should always go back to the original document, not trust the AI's summary. A quick trick: ask yourself, 'Would I bet my job on this citation being real?' If not, verify it.

A compact way to remember this: 'AI can cite ghosts.' Imagine a ghost holding a fake diploma—it looks official but is not real. When you see a citation from AI, treat it like that ghost: check its credentials before relying on it. This mental image helps you recall that citations are not proof, and verification is the only safe step.