What to Read

Grounding means connecting an AI answer to specific, verifiable sources. RAG is a common implementation pattern: retrieve relevant information, then generate using that context.

Exam shortcut: if the question says "official docs only", "company policy", "latest source", or "reduce hallucination", think grounding or RAG.

Without vs With Grounding

Dimension	Model only	Model plus RAG
Knowledge source	Training data and prompt context.	Retrieved docs, databases, search systems, or enterprise sources.
Freshness	Can be stale.	Can use current retrieved knowledge.
Trust	Harder to verify.	Can cite or trace source material.
Hallucination risk	Higher when facts are missing.	Lower when generation is constrained to retrieved context.

Prompting Patterns

Zero-shot: ask directly without examples.
One-shot: provide one example before the task.
Few-shot: provide multiple examples to shape style or format.
Role prompting: ask the model to act as a specific expert or persona.
Prompt chaining: split a workflow into smaller prompts where each output feeds the next step.
ReAct: combine reasoning with actions or tool calls.

Practical Use Cases

Need	Likely pattern
Answer questions over internal policy documents.	RAG or NotebookLM-style grounding.
Make brand-consistent marketing copy.	Role prompting plus few-shot examples.
Summarize long uploaded PDFs.	NotebookLM or source-grounded assistant.
Debug a multi-step workflow.	Prompt chaining with clear intermediate outputs.