Build and maintain a Knowledge Base

Plan the knowledge before uploading

Organize content around services, prices, locations, opening hours, eligibility, process, policies and frequently asked questions. Remove outdated or contradictory material before training.

Add text and files

Use Text for clean, frequently updated business information. Use Files for supported documents that contain useful customer-facing knowledge.

• Use descriptive headings and short sections.
• Prefer selectable text over image-only documents.
• Remove internal notes, credentials and customer information.
• Avoid uploading the same information in conflicting versions.

Use structured question-and-answer pairs

Q&A pairs are useful when a question requires a precise approved response. Add pairs manually or use the supported CSV import. Export existing pairs before a large update so you retain a recoverable reference.

Train from website sources

Enter approved URLs under Websites. Decide whether to fetch only the supplied URL or follow internal links. Advanced options can control recurring fetches, header removal, main-content extraction and specialized fetching.

• Fetch only public pages you are authorized to use.
• Exclude legal boilerplate, navigation duplication and irrelevant blog archives when possible.
• Review the fetched URL list, character count and status.
• Re-fetch changed pages and remove obsolete sources.

Search and verify retrieval

Use Knowledge Base Search with realistic customer questions. Confirm that the most relevant information appears. Then test the same question in the live agent because retrieval and final response generation are separate steps.

Maintain quality

1. Review unknown-answer alerts and conversation transcripts weekly.
2. Correct the source information, not only the final prompt.
3. Re-train after meaningful content changes.
4. Test prices, policies and booking instructions after every update.
5. Track who approved sensitive customer-facing information.