The correct choice is Retrieval Augmented Generation (RAG). This approach anchors assistant replies to trusted company information by fetching authoritative sources at inference time.

By retrieving the live product catalog, manuals, and policy documents when generating answers Retrieval Augmented Generation (RAG) grounds responses in verifiable facts which reduces hallucinations and keeps answers current.

Lower temperature may make wording more consistent but it does not add missing knowledge and cannot guarantee factual accuracy.

Fine-tuning can help with style and domain adaptation but it bakes knowledge into the model making updates harder and it can still hallucinate for new or changing product data.

Guardrails help enforce safety and policy boundaries but they do not retrieve or verify facts so they do not prevent incorrect product claims.

Full AWS Practitioner Certification Question