Retrieval augmented generation (RAG)
Retrieval augmented generation (RAG) is a technique in which an AI language model first retrieves relevant information from its own knowledge source — such as documents, manuals or a database — and bases its answer on it.
Language models only know what was in their training data. With RAG they answer questions based on current, company-internal information — and can cite the source.
This significantly reduces the risk of invented answers, but does not eliminate it entirely. Quality depends heavily on how well the knowledge source is maintained and prepared.
For SMEs, RAG suits internal knowledge assistants, customer service or searching contracts and manuals — for sensitive data with operation in Germany and clear access rights.
What to look out for
- Select and clean the knowledge sources: current, consistent documents instead of the file share as-is.
- Respect access rights: each user should only get answers from documents they are allowed to see.
- Return answers with source references so they can be verified.
- Define how new or changed documents are added.
- Test quality regularly with typical questions.
Common mistakes
- Outdated and current versions of the same document sit side by side — the AI quotes the wrong one.
- Answers without sources that nobody can verify.
- Confidential documents become searchable for all employees.
Frequently asked questions
It significantly reduces the risk because answers are based on concrete documents, but does not eliminate it completely. Source references and spot checks remain important.
No. RAG uses an existing language model and connects it to your own documents — no custom training is required.
More in our article: AI for SMEs: 7 applications that already save money today