Chapter 09 · Sustainability & AISustainability & AI
Retrieval-augmented generation (RAG)
Definition
An architecture that grounds a language model's outputs in documents retrieved at query time, so answers draw on specified sources rather than model memory alone.
References
This source provides part of the technical or institutional basis for the definition.
This source supports the explanation of how the term is applied, measured or governed in practice.
Overview
What it means
Introduced by Meta AI researchers in 2020 for knowledge-intensive tasks, RAG pairs a search index with a generator: the system fetches relevant passages, then writes an answer conditioned on them. This makes sources inspectable and updatable without retraining.
How it is used
RAG is the standard pattern for AI tools over corporate sustainability knowledge: querying a company's own reports, regulatory libraries or product data with citations. It reduces — but does not remove — the risk of unsupported statements, since retrieval can miss or generation can distort.
Why it matters
For evidence-based fields, RAG changes AI from an oracle into a research assistant that shows its sources. That traceability is what makes generative AI usable in defensible sustainability analysis and disclosure work.