Retrieval augmented generation (RAG) reduces AI hallucinations by anchoring responses to real documents, aiming to minimize mistakes. The method is described as ‘low hallucination’ but not entirely error-free, with ongoing debate on its effectiveness.