What is retrieval-augmented generation?

It is when an AI model pulls live documents from a search index or the web while it is answering, then writes its reply from that retrieved text. It lets the model cite recent or specific sources instead of guessing from training data.

Does RAG mean my page will be used in answers?

No. RAG is the mechanism, not a promise about your site. Your page has to be retrievable and readable for a model to pull it, and even then the engine decides what to use. You can measure whether your page is readable, not guarantee it gets pulled.

How is RAG different from a model just knowing things?

A plain model answers from memory frozen at training time. A RAG system fetches fresh documents at the moment of the question, so the answer can reflect pages it never saw during training, including yours.

← Back to glossary

Glossary · 2 min read

Retrieval-augmented generation (RAG)

A method where an AI model fetches outside documents at answer time and writes its reply from what it just read, instead of relying only on what it memorized during training.

You asked an AI a question and it answered with a fact from this morning. A model frozen at training time cannot do that. Something fetched a live page first.

That something is retrieval-augmented generation, and it is the reason your website can end up inside an answer at all.

What RAG actually does

Retrieval-augmented generation is a two-step move. First the system retrieves documents that look relevant to the question. Then the model generates its reply from what it just read.

The model is not answering from memory alone. It is reading fetched text in real time and writing on top of it. When an engine quotes a price, a date, or a definition it could not have memorized, retrieval is why.

Your page is one of the documents that step one can pull. If it gets pulled and read cleanly, it can shape the answer.

Old way, new way

The old way: a search engine retrieves ten links and a human clicks one to read your page. Retrieval ended at the link.

The new way: the engine retrieves your page, reads it itself, and writes the answer. Retrieval now ends inside the model, not on your site. The reader you are writing for changed from a person to a parser.

What decides whether yours gets pulled

Two gates sit between your page and a RAG answer.

The first is retrievability. A crawler has to be able to reach and index the page. If it is blocked, slow, or invisible to a fetch, it never enters the candidate pool.

The second is readability. Even a retrieved page is useless to the model if the answer is buried under layout, the heading lies about the content, or nothing labels what the page is. The model reads what it can parse first and may never reach a point stranded far down the page.

The damaging admission

Citedon does not run the retrieval. We do not sit inside ChatGPT, Perplexity, Gemini, or Claude, and we cannot make any of them fetch your URL. No tool can.

What we can do is measure the part you control: whether your page is reachable and readable enough to survive both gates if a model does reach for it. The engine's choice stays the engine's choice.

How to check yours

Pick the page you would most want an engine to pull when someone asks about your business. Ask a blunt question: if a model retrieved this exact URL right now, could it parse a clear answer, or just a wall of layout?

Run a free scan on that URL to see how a machine reads it today, or read how AI engines decide what to recommend for the longer picture.

See whether engines can read the pages they would retrieve, free.

Run a free scan. No signup. You get a readiness score and the gaps to fix, in about a minute.

Run a free scan How it works