Retrieval-augmented generation

What it is

Retrieval-augmented generation is an approach where an LLM fetches relevant documents at query time and grounds its answer in that retrieved material rather than relying solely on its trained weights. It is the pattern behind most enterprise AI search products and Perplexity-style answer engines.

Why it matters

It determines whether and how your content is selected, quoted, and attributed in AI answers, making retrievability and clear, chunk-level structure as important as traditional ranking.

How it works

A system embeds and indexes source content, retrieves the closest passages to a query, and passes them as context so the model composes a grounded, often cited response.

When it applies

It applies wherever answers must reflect current, source-specific, or proprietary information rather than the model's static training data.

Examples

• An answer engine retrieving three articles and synthesising a cited summary
• An internal support bot grounding replies in a company's documentation index
• A site optimising self-contained sections so passages retrieve cleanly out of context

How it is measured

• Retrieval relevance of passages returned for a target query
• Faithfulness of generated answers to the retrieved sources
• Citation accuracy and frequency for your content in answers
• Coverage of expected queries by retrievable, well-structured passages