# Grounding, Citations and Saying Not Found — Prompt & Context Engineering

Source: https://www.skillbyai.com/en/prompt-context-engineering/x-ground

> Make answers traceable to the context and honest about gaps.

## Answer from the sources or say so

For knowledge tasks, instruct the model to answer **only from the provided documents**, to **cite** the document IDs it used, and to reply with a clear **not found** when the documents do not contain the answer. Give documents stable IDs in the context so citations can be checked in code (does the cited ID exist, and does the quoted text appear in it?). Test with questions whose answers are deliberately missing: a system that never says not found is guessing.

## A grounded answer format

Citations make answers checkable.

```text
Instructions: Answer only from the documents. Cite ids like [doc-2].
If the answer is not in the documents, reply exactly: NOT_FOUND.

Example answer:
Refunds are issued within 14 days of receiving the item [doc-1].
The customer pays return shipping [doc-3].
```

## Test unanswerable questions

Include questions with no answer in the documents in your evaluation and measure how often the system correctly says not found.

**Quiz:** Why include unanswerable questions in tests?

- [ ] Because models cannot cite
- [ ] To make tests faster
- [ ] To lower the token count
- [x] To check the system admits missing information instead of guessing

*Answer:* To check the system admits missing information instead of guessing. Honest not-found answers are a quality signal.