What is Retrieval-Augmented Generation (RAG)?
Retrieval-augmented generation (RAG) is a method in which an AI model first looks up relevant information, such as web pages or documents, and then writes its answer based on what it found. Instead of relying only on what it memorized during training, the model works with fresh, specific sources.
A RAG system works in two steps. First, retrieval: the system searches an index (the web, a search engine or a company database) for the passages that best match the question, often using embeddings to compare meaning rather than exact words. Then generation: those passages are handed to a large language model, which writes the answer and can point to its sources.
The term comes from a 2020 research paper. Today the same idea sits behind AI assistants that search the web before answering, such as Perplexity, ChatGPT search and Google AI Overviews.
Why RAG matters for AI visibility
RAG is the main reason your current website content can influence AI answers today, not only after the next model is trained. If your page is retrieved for a prompt, the model can quote your facts, mention your brand and link to you as a source. If it is not retrieved, you depend on whatever the model remembers from older training data.
It also changes what competition looks like. The model works with a limited set of retrieved passages per question, so you are competing to be one of the few sources it actually reads. That is how an AI citation is earned: your page was retrieved, read and judged useful enough to support the answer.
How to optimize your content for RAG
Be retrievable. Let search engine and AI crawlers access your pages, serve the main text in the HTML and keep important pages indexed in Google and Bing.
Write passages that stand on their own. Retrieval often picks a paragraph, not a whole page, so each section should answer one question clearly.
Put facts first. Lead with the answer, then add detail. Include specifics such as prices, features, locations and dates.
Get covered elsewhere. Reviews, comparisons and trade media are retrieved too. Our guide on how to get cited by AI explains where to start.
For the first step, Trackbase Crawler Analytics shows which AI crawlers visit your site and which pages they read.
Examples of retrieval-augmented generation
A user asks Perplexity which CRM suits a small agency. It searches the web and reads several comparison articles and vendor pages. Then it writes a shortlist with numbered source links.
Google AI Overviews answers "how long does a heat pump last" with a summary drawn from several web pages, with links to those sources.
A B2B payroll software vendor publishes a clear pricing page. Once it is indexed, assistants that search the web can quote the new prices, while a model answering from memory still gives the old ones.
A Dutch bicycle webshop connects its help center to a customer service chatbot, so the bot answers delivery questions from the shop's own policies instead of guessing.
Frequently asked questions
What is the difference between RAG and training data?
Training data is what a model learned from before its release, and that knowledge stops at the knowledge cutoff. RAG adds information at the moment of the question by retrieving current pages or documents. In practice, training data shapes what a model already believes about your brand, while RAG decides which live sources it reads and cites when it answers.
Do ChatGPT and Perplexity use RAG?
Yes, whenever they search the web. Perplexity searches for almost every question and shows its sources. ChatGPT searches the web when a question needs current information or when the user asks it to. In both cases the assistant retrieves pages first and writes the answer from them, which is why well-structured, crawlable content can appear in their answers.
How can I tell if my content is used in RAG answers?
Look for citations. When an assistant links to your page as a source, your content was retrieved and used. You can also check server logs or crawler analytics for user-triggered bots such as ChatGPT-User and Perplexity-User, which fetch pages when someone asks a question. Tracking prompts over time shows how often your pages are cited.