What is LLM hallucination?

An LLM hallucination is when the model states something false as if it were true — confidently inventing facts, sources, names, or numbers that simply aren't real. It's not lying, because the model has no intent; it's a side effect of how these systems work. An LLM generates text by predicting what sounds plausible, not by looking up verified facts.

So when it doesn't actually know something, it often fills the gap with a fluent, convincing guess.

Why do LLMs hallucinate?

The root cause is that an LLM predicts plausible-sounding text rather than checking a source. It has no built-in database of verified facts and no real sense of what it does and doesn't know. When a question falls outside its training, involves very recent events, or asks for precise details like dates and figures, the model still produces a confident answer — because generating fluent text is exactly what it was built to do.

The tricky part is the tone: a hallucination usually sounds just as authoritative as a correct answer, which makes it easy to miss.

What do hallucinations look like?

They show up most often when specific, checkable details are involved:

• Citing studies, articles, or court cases that don't exist.
• Inventing a quote and attributing it to a real person.
• Misstating a statistic, date, or figure.
• Describing a product feature that was never built.

What these share is specificity: the model reaches for a concrete-sounding detail to finish the answer, and when it doesn't have a real one, it manufactures something that fits.

How can you reduce hallucinations?

You can't fully eliminate hallucinations, but you can make them far less likely. Grounding the model in real sources — the core idea behind retrieval-augmented generation — lets it answer from actual documents instead of memory. Asking it to cite evidence, and then checking those citations, catches many invented facts.

And for anything high-stakes, keep a human in the loop. The key mindset: treat an LLM as a fast, capable, but occasionally unreliable assistant — a reason to verify important facts, not to blindly trust a confident answer.