Retrieval-Augmented Generation (RAG)
Retrieval-augmented generation (RAG) is the technique where an AI model first searches for relevant documents, such as web pages or listings, and then writes its answer from what it retrieved, instead of relying only on what it memorized during training.
Also known as: RAG
Retrieval-augmented generation is a two-step process. First, retrieval: the system takes your question, runs a search across an index of web pages, documents, or listings, and pulls back the most relevant passages. Second, generation: the language model reads those passages and writes an answer grounded in them, often citing the sources it used. Nearly every answer engine that names real businesses works this way, because a model's training data alone doesn't know which electrician in Boise is open this week.
The retrieval step is where your website comes in. When someone asks ChatGPT for an estate planning attorney in Raleigh, the model doesn't recall your firm from training. It searches, and your pages either come back in the results or they don't. Ahrefs found that 88% of the URLs ChatGPT cited across 1.4 million prompts came from its general web-search retrieval rather than its news, Reddit, YouTube, or academic sources. Retrieval is the main door, and your pages have to be findable through it.
RAG explains a lot of the strange behavior people notice in AI answers. The same question asked twice can retrieve slightly different documents and produce a different list of names. A business with a great reputation but a thin website can be skipped because there's little for retrieval to find. And retrieval doesn't simply mirror Google: Ahrefs found only 12% of AI-cited URLs ranked in Google's top 10 for the original prompt, so ranking well is no guarantee of being retrieved.
For a small business, the practical takeaway is that the model is only as good as what it can fetch about you. Pricing stated in plain text, a service list, your service area spelled out, and consistent facts across your site and directories all give retrieval something concrete to return. A PDF menu, an image of your price list, or a site that blocks AI crawlers gives it nothing.
Tibly's cited-source tracking shows which pages and directories each engine actually retrieved when it answered a question in your trade and city. That's the retrieval half of RAG made visible: you can see whether your own site came back, or whether the engine built its answer from Yelp, a competitor's page, and a Reddit thread instead.
Keep exploring the glossary
- GroundingGrounding is when an AI model bases its answer on live, retrieved information, such as a web search result, instead of relying only on what it learned during training, which is meant to make answers more current and accurate.
- Cited SourceA cited source is a specific web page or domain, such as a Yelp listing, a Reddit thread, or a business's own service page, that an AI engine references as the basis for its answer, showing where it learned it.
- Answer EngineAn answer engine is an AI system, such as ChatGPT, Perplexity, Gemini, Google's AI Overviews, or Copilot, that responds to a question with a direct written answer naming a small number of options instead of a page of ranked links.
- Hallucination (AI)A hallucination is when an AI system states something as fact that is false, unsupported by any source, or not actually said by the source it appears to cite, presented with the same confidence as a correct answer.
Common questions about Retrieval-Augmented Generation (RAG)
- Is RAG the same as grounding?
- Close. Grounding is the goal: tying an answer to real sources so it's accurate. RAG is the most common technique for achieving it, by retrieving documents before generating. In practice people use the terms interchangeably when talking about how AI engines source their answers.
- If my business isn't in the model's training data, can it still recommend me?
- Yes, and that's the point of RAG. The model searches the web at answer time, so a two-year-old practice with a clear, crawlable website can be retrieved and named. What matters is whether your pages come back in that search and state the facts the model needs.