A knowledge base is the source material an answer is drawn from. For people it might be a help center or internal wiki. For an AI system it is the set of documents that Retrieval-Augmented generation searches to AI agent grounding its answers.
The quality of the answers can never be better than the quality of the knowledge base. Content that is outdated, contradictory or duplicated will be retrieved and repeated. Keep it current, give each document a clear owner and date, and remove what is no longer true.
For AI use, the content is usually split with document chunking, turned into embeddings and stored in a vector database. Well-written, self-contained pages with clear headings tend to work better than long, rambling ones.
Related: Semantic Search, AI Agent, LLM Hallucination.