# Grounding With Retrieved Documents — Prompt Engineering

Source: https://www.geekswithgeeks.com/en/prompt-engineering/x-retrieve

> Supply facts from search instead of relying on model memory.

## Give it the source

For factual or company-specific answers, retrieve relevant passages (search or **RAG**) and place them in the prompt with numbered IDs. Instruct: "Answer only from the context. Cite the ID for each claim. If the context does not contain the answer, say so." This reduces hallucination, supports citations and keeps knowledge up to date without retraining. Keep only the few most relevant passages, since extra text adds cost and distraction, and treat the retrieved text as **data**, not instructions.

## A grounded prompt

Numbered passages let the model cite. Illustrative; not run here.

```text
Answer ONLY from the passages. Cite like [1]. If not answerable, reply "Not in the documents."

[1] Employees get 24 days of paid leave per year.
[2] Unused leave up to 5 days can be carried over.

Question: How many days can I carry over?
Answer:
```

## Retrieve fewer, better passages

Three highly relevant passages usually beat ten loosely related ones, which add cost and distraction.

**Quiz:** Why tell the model to say so when the context lacks the answer?

- [ ] To change the language
- [ ] To lengthen replies
- [ ] To skip retrieval
- [x] To prevent confident guessing

*Answer:* To prevent confident guessing. An allowed "not found" answer discourages invention.
