# Grounded Answers From Retrieved Context — Dify Workflow Basics

Source: https://www.skillbyai.com/en/dify-workflows/k-rag

> Pass context to the LLM node and keep it honest.

## Context in, citations out

A typical RAG workflow: start (query) to knowledge retrieval (chunks) to LLM node (instructions, context, question) to end. Instruct the model to answer **only from the context**, to say when the answer is not there, and to cite sources. Enable citation or source display where available so users can check answers. Combine with the threshold so off-topic questions receive "I don't know" rather than an answer built on an unrelated chunk. Dify's node names, menus and options change between versions; check the current Dify documentation.

## A grounded answer prompt

Put context and question in clearly marked sections.

```text
System:
You answer customer questions using ONLY the context below.
If the context does not contain the answer, reply: "I could not find this in our help articles."
Quote the article title you used.

Context:
{{#knowledge_retrieval.result#}}

Question:
{{#start.query#}}
```

## Test unanswerable questions

Include questions your documents cannot answer in testing; a good RAG app admits it rather than inventing.

**Quiz:** What should a grounded RAG app do when the context lacks the answer?

- [ ] Return the whole knowledge base
- [ ] Invent a plausible answer
- [x] Say it could not find the answer
- [ ] Crash

*Answer:* Say it could not find the answer. Honest refusals build trust.