Most AI answer engines build a reply in two steps: they retrieve relevant sources, then write an answer based on them. That two-step method is called retrieval-augmented generation, or RAG. It matters for your brand because the answer is only as good as the sources the AI retrieves. Improve what it can find about you, and you improve the answer. Here is how it works.
The two kinds of knowledge
An AI tool draws on two things when it answers.
The first is training data: the large body of text the model learned from before it answered any question. This gives it general knowledge, but it is fixed until the model is retrained, and you cannot edit it directly.
The second is live retrieval: for many questions, the tool searches the web in real time, pulls in current pages, and uses them to answer. This is the part that changes daily and the part you can influence.
Every major engine now works this way. ChatGPT decides on its own when to search the web and links to the sources it used (OpenAI, Introducing ChatGPT search). Claude searches the web and returns cited responses on all plans (Anthropic, Claude can now search the web). Perplexity and Google AI Overviews were built around live retrieval from the start.
What RAG actually means
Retrieval-augmented generation (RAG) is the name for that second method. Break the phrase down: the model retrieves real documents, then generates an answer augmented by, meaning grounded in, those documents.
In plain terms: instead of answering purely from memory, the AI first goes and finds relevant, current sources, then writes its answer from them. This is why these tools can cite links and stay reasonably up to date.
How it picks what to retrieve
The AI does not grab every page. It pulls a set of candidate sources and favors the ones it trusts. A few things push a source into that set:
- Relevance: it clearly answers the specific question asked.
- Consensus: many independent sources say the same thing, so it reads as reliable.
- Authority: the platform already trusts the site or community.
- Freshness: for many questions, recent content beats old content.
Some engines also break your question into several smaller searches behind the scenes, gather sources for each, and combine them. That is called query fan-out. Google confirms its AI Overviews and AI Mode use this technique, issuing multiple related searches across subtopics to build one response (Google Search Central, AI features and your website). It is why one answer can pull from many different sites, and why pages that never rank for the literal question can still get cited.
Why this matters for your brand
If the answer is built from retrieved sources, then your visibility depends on what the AI can retrieve about you. Thin or out-of-date sources mean a thin or wrong answer. Credible, accurate, current sources mean a better one.
This is also why you cannot just tell the model your brand is great. It answers from evidence it retrieves and can check, so the way to change the answer is to improve that evidence on the sources it reads.
Where to go next
For where those sources come from, read what is AEO, GEO and AI search. For the terms used here, see the AEO glossary.
