Retrieval-augmented generation
Retrieval-augmented generation is a technique where an AI system first retrieves relevant passages from your own documents and then answers using them, so the answer is grounded in sources you can check rather than in the model's memory.
Also written: RAGArabic: التوليد المعزّز بالاسترجاع
The problem it solves
A general model does not know your policies, your contracts or your prices, and asked anyway it will produce something confident and wrong. RAG changes the question from "what do you know" to "what do these documents say", which is a question a model is far better at.
What it does not solve
RAG reduces invented answers; it does not eliminate them. If retrieval returns the wrong passage, or nothing, the answer degrades. This is why a usable system shows its sources — so the reader can check the answer rather than trust it.
What makes it work in practice
Less the model than the plumbing: which documents are in scope, how they are split, how the system behaves when it finds nothing, and whether answers carry citations. A system that says "I could not find that" is more valuable than one that always produces an answer.
Questions people ask
Only if you choose an architecture where they do. Retrieval can run on your own infrastructure, and which model sees what is a decision to make deliberately.
You check the cited sources. Any deployment worth trusting also has an evaluation set — real questions with known answers — that is re-run when anything changes.
Related serviceAI for Business