Hallucination occurs because LLMs generate text based on statistical patterns, not factual knowledge. The model produces what 'sounds right' based on training data, which sometimes means inventing citations, making up statistics, or confidently stating falsehoods.
Mitigation strategies: RAG (ground responses in retrieved facts), chain-of-thought (force explicit reasoning), tool use (let the model look things up), and human verification. Hallucination is not a bug to be fixed — it's a fundamental property of probabilistic language generation.