What it means for a UK law firm
Retrieval-augmented generation is the usual way legal AI tools answer questions about a firm's own documents. The retrieval step decides which passages the model sees. If the right clause is not retrieved, the answer can be incomplete even when every citation in it is accurate.
Retrieval-augmented generation was set out in a 2020 research paper by Lewis and others. The term is now used for any system that combines a search step with a language model.
In legal work, “RAG” also commonly means red, amber, green risk ratings. The two meanings are unrelated; see document risk assessment.
Example
Illustrative scenario
A firm asks its AI tool which leases in a client's portfolio contain a landlord's break clause. The tool retrieves the break clauses it finds and answers from them. The solicitor checks the answer against the schedule of leases, because a lease the search step missed will not appear in the answer at all.
How Chancery handles it
Chancery reads each uploaded document, including scanned PDFs, and makes the whole document searchable. Questions can be scoped to one matter, all of your matters or your precedent library, and every document answer cites the passage it relies on; see Matters.