RAG (Retrieval-Augmented Generation)

Retrieval-augmented generation is when an AI searches a set of documents first, then writes its answer from what it found. It is how AI tools answer questions about current events or private files that were never in their training data.

It runs in two steps. A retrieval system finds the passages most relevant to the question, then hands them to the language model as context, and the model writes from them. Perplexity works this way. So does Google’s AI Overviews, and so does any internal chatbot answering questions about a company’s own files. RAG is the reason answer engines can cite sources at all: the citations are the retrieved documents.

This has a direct consequence for how you write. RAG systems do not retrieve whole pages, they retrieve chunks of a few hundred words. A chunk that depends on earlier context to make sense will retrieve poorly and read badly when it does. Sections that stand alone win.

For nonprofits that argues for one specific discipline: every section under a heading should make sense to someone who lands on it cold. Define acronyms on each use. Repeat essential context rather than referring back to it.

Related reading: Your rankings held. Your traffic didn’t.

What this means for your organisation

Open your most important report and read one middle section in isolation. If it is incomprehensible without the preceding pages, it will never be retrieved usefully.

Related terms