LearnAI

Navigation

Introduction to AI

Machine Learning

Deep Learning

NLP & LLMs

Computer Vision

Generative AI

Reinforcement Learning

AI Ethics & Governance

Tools & Frameworks

General

Introduction to Retrieval Prompting Patterns

Retrieval Prompting Patterns are standardized ways of formatting how retrieved knowledge is presented to a large language model (LLM) within a Retrieval-Augmented Generation (RAG) system.

  1. Context-First: The retrieved information is placed at the top of the prompt to provide the model with “grounding” before the instruction.
  2. Context-Instruction-Query: A three-part structure used to clearly separate knowledge from the user’s specific request.
  3. Few-Shot RAG: Providing examples that include both the retrieved background and a high-quality demonstration answer.

Key Principles

  • Clarity: Use delimiters (e.g., ### Context ###) to help the model distinguish between your instructions and the retrieved data.
  • Relatability: Explicitly tell the model “Use ONLY the provided context” to reduce hallucinations.
  • Relevance: Only include the most relevant chunks to avoid distracting the model with unrelated information (“Lost-in-the-Middle” problem).

Common Prompt Template example

Use the following pieces of context to answer the question at the end.
If you don't know the answer, just say you don't know.

Context:
{context}

Question: {question}
Helpful Answer: