In prompt design, what practice helps prevent hallucinations when using background knowledge retrieval?

• A. Use retrieved results with verification; source selection; freshness checks; and verifying with retrieval results.
• B. Avoid using retrieved results entirely.
• C. Use up-to-date facts and do not cite sources.
• D. Rely on internal facts and ignore retrieval.

Answer: (A)

When designing prompts, the idea is to ground the model’s output in external information while keeping it checked and trustworthy. Pulling in background knowledge is powerful, but retrieval by itself can lead to mistakes if the model misreads a source or fills gaps with fabrication. Verification acts as a safeguard: it checks that the model’s claims align with what was retrieved. Verifying with retrieval results means cross-referencing the generated content against the actual retrieved material and, if needed, citing exact sources or passages.

Source selection matters because not all sources are equally reliable or relevant; choosing credible, appropriate sources reduces the chance of spreading false information. Freshness checks ensure the information isn’t outdated, which is crucial for topics that change over time. Put together, these practices create a loop: retrieve, verify, select trustworthy sources, ensure up-to-date facts, and re-check alignment with the retrieved content.

Other approaches—avoiding retrieval, ignoring sources, or not citing sources—remove the anchoring that keeps outputs grounded, increasing the risk of hallucinations.