---
title: "Discussion on \"Why Does My RAG System Still Give Wrong Answers? A Beginner's Guide to Retrieval-Augmented Generation\" | Hashnode"
description: "Discussion on \"Why Does My RAG System Still Give Wrong Answers? A Beginner's Guide to Retrieval-Augmented Generation\". If you've ever built a chatbot on top of an LLM and asked it something about your own company's documents, you've probably hit this wall:\n\n\"I don't have information about that.\"\n\nOr worse — it makes s"
url: "https://hashnode.com/posts/why-does-my-rag-system-still-give-wrong-answers-a-beginner-s-guide-to-retrieval-augmented-generation/6a578282ec970043f3231172"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Why Does My RAG System Still Give Wrong Answers? A Beginner's Guide to Retrieval-Augmented Generation](https://anshu1.hashnode.dev/why-does-my-rag-system-still-give-wrong-answers-a-beginner-s-guide-to-retrieval-augmented-generation?utm_source=hashnode&utm_medium=feed)

If you've ever built a chatbot on top of an LLM and asked it something about your own company's documents, you've probably hit this wall: "I don't have information about that." Or worse — it makes s

Jul 15 · 7 min read · anshu1.hashnode.dev  
Read the article

-   [#rag](https://hashnode.com/tag/rag)

## Written by

[Anshuman Nayak](https://hashnode.com/@anshumannayak)

@anshumannayak · odisha,india

  
[View profile](https://hashnode.com/@anshumannayak)

Building intelligent software with AI, Full-Stack development, and modern cloud technologies.

Building scalable web applications and AI-powered solutions. Sharing practical tutorials, project insights, and lessons from real-world development.

## Responses 1

Markdown supported · ⌘↵ to post

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[Jul 16](https://hashnode.com/posts/why-does-my-rag-system-still-give-wrong-answers-a-beginner-s-guide-to-retrieval-augmented-generation/6a578282ec970043f3231172/comment/6a58e07f96e1a6bb88648a63)

The open-book versus closed-book framing is the clearest way I have seen to explain why retrieval quality caps answer quality regardless of model strength. In practice most RAG wrong answers I see trace back to chunking and retrieval recall, not generation, so I score retrieval hit rate before blaming the LLM. Do you evaluate retrieval and generation as two separate metrics, or judge the final answer end to end?
