Skip to content

Comparison

RAG versus a raw language-model prompt

A model can write fluent language. An enterprise document workflow also needs retrieval, evidence someone can check and rules about who sees what.

Definition

RAG supplies retrieved material to the model; it does not make the model inherently correct or private.

01

A raw prompt can fit

Use it for low-risk drafting or transformation when the user supplies the relevant context and provider terms meet the requirement.

  • Low-risk drafting
  • Text transformation
  • User supplies the context
  • Provider terms acceptable
02

A governed retrieval workflow can fit

Use it when sources change, access differs by user, evidence must be inspected or the same test needs to run repeatedly.

  • Document set control
  • Passage retrieval
  • Citation resolution
  • Evaluation and traces

What this page does not prove

  1. B1RAG can retrieve irrelevant material.
  2. B2A citation can be present and still fail to support the claim.

Related pages

Plan an evaluation

Test the claim on your documents

Pick a real piece of work, agree what a good answer looks like, then go through the results together.

Book a demoExplore the platform