Glossary

What is retrieval-augmented generation (RAG)?

Retrieval-augmented generation (RAG) is an approach that retrieves relevant information and supplies it to a generative model when producing an answer. It can connect an AI application to product records, documents or other knowledge sources. RAG adds evidence to the generation process; it does not guarantee that retrieval or the resulting answer is correct.

Retrieval before generation

A RAG workflow searches an available knowledge collection, selects relevant material and includes that material in the model's context. The model then produces a response using the question and supplied information. This can make domain-specific or recently updated material available without retraining the underlying model for each change.

For example, a technical support assistant could retrieve the correct product manual and approved catalogue record before answering a compatibility question. The application should retain references so a user can inspect the evidence behind the answer.

The retrieval step can fail too

A similar product name can lead to the wrong document or variant. A source may be outdated, incomplete or contradictory. The model can also misinterpret retrieved text or add an unsupported statement. Checking that some source was retrieved is therefore insufficient.

Evaluate whether the workflow finds the correct product, retrieves the required passage and produces an answer supported by it. Keep document versions and product identifiers connected. RAG is a way to supply context and support grounding, not a replacement for source quality, access control or explicit handling of missing evidence.

Related terms

Sources and further reading

AI-ready product data

Discuss the product information and controls your AI use case needs.