context injection

Retrieval has found the passages; now they have to actually reach the model. Context injection is the step that pastes the retrieved text into the prompt — typically under a heading such as "Context:" followed by the passages, then your question and an instruction to answer using that context. It is the bridge between the search half of RAG and the generation half.

How you assemble this matters more than it looks. The passages must fit inside the context window, so you keep only the top few. Their ordering, clear separators between them, and a source label on each all change how reliably the model uses them. Placing the most important material near the start or the end helps too, because models tend to neglect the middle of a long prompt. The model then generates its answer conditioned on this injected evidence.