Start with source quality, not source quantity
A large knowledge base can still fail if documents are outdated, contradictory, or written for internal teams instead of customers.
Prioritize clean, public-facing sources first: help center docs, pricing explanations, onboarding guides, and policy pages.
Structure content for retrieval
Bots retrieve small text chunks. Dense pages with many unrelated topics create noisy matches and weak answers.
- One topic per section with clear headings.
- Short paragraphs and explicit examples.
- Keep dates, limits, and policy versions visible in text.
- Avoid screenshots for critical information.
Common mistakes to avoid
Many teams upload old docs and then forget to remove them. The bot then picks stale answers with high confidence.
- Do not upload duplicate documents with different wording.
- Do not mix internal SOPs with customer instructions in one source.
- Do not leave deprecated feature docs active.
Maintenance cadence
Review top failed questions weekly and patch missing coverage immediately. This one habit drives the biggest quality gains over time.