Retrieval: Definition and Examples

Retrieval, or information retrieval, is a fundamental mechanism in artificial intelligence that allows a language model to access external knowledge when generating a response. Rather than relying solely on what it learned during training, the model queries a structured data source to find the most relevant passages for the question asked.

This process typically relies on semantic search techniques: the user's query is transformed into a numerical vector (embedding), then compared to vectors of documents stored in a vector database. The semantically closest documents are then selected and injected into the model's context before response generation. This is the principle at the heart of the RAG (Retrieval-Augmented Generation) architecture.

The major benefit of retrieval is enabling AI models to provide up-to-date, factually grounded, domain-specific responses without costly retraining. A company can thus connect an LLM to its internal documentation, knowledge bases, or archives, and obtain contextualized and reliable answers.

The quality of retrieval directly determines the quality of the final response. Poor retrieval — irrelevant, too many, or poorly segmented documents — leads to hallucinations or off-topic answers. This is why optimizing the retrieval pipeline (chunking, ranking, filtering) has become a discipline in itself in AI systems engineering.